Skip to content

Instantly share code, notes, and snippets.

@davidberard98

davidberard98/67434_0_log.out

Last active September 23, 2022 00:49
Show Gist options
  • Save davidberard98/fc12ac6b42b4e2ab7ad5a1711797966d to your computer and use it in GitHub Desktop.
Save davidberard98/fc12ac6b42b4e2ab7ad5a1711797966d to your computer and use it in GitHub Desktop.
Download ZIP
hf_T5, 2 nodes, dynamo+inductor, verbose=True, log_level=DEBUG; functorch..debug_graphs is FALSE.
Raw
67434_0_log.out
This file has been truncated, but you can view the full file.
submitit INFO (2022-09-22 18:42:53,293) - Starting with JobEnvironment(job_id=67434, hostname=a100-st-p4d24xlarge-3, local_rank=0(8), node=0(2), global_rank=0(16))
submitit INFO (2022-09-22 18:42:53,294) - Loading pickle: /fsx/users/dberard/scratch-local/bench-fast/benchmark/logs/67434_submitted.pkl
Process group: 16 tasks, rank: 0
MY HOSTNAME: a100-st-p4d24xlarge-3
FI_PROVIDER : efa
LD_LIBRARY_PATH : /fsx/users/dberard/scratch-local/bench-fast/aws-ofi-nccl/lib:/opt/amazon/efa/lib:/fsx/users/dberard/scratch-local/bench-fast/aws-ofi-nccl/lib:/opt/amazon/efa/lib:/path/to/aws-ofi-nccl:/opt/amazon/efa/lib:/path/to/aws-ofi-nccl:/opt/amazon/efa/lib:/usr/local/cuda-11.6/lib:/usr/local/cuda-11.6/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/usr/local/cuda/efa/lib:/usr/local/cuda/lib:/usr/local/cuda:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64:/usr/local/cuda/targets/x86_64-linux/lib:/usr/local/lib:/usr/lib:
NCCL_DEBUG : TRACE
FI_EFA_USE_DEVICE_RDMA : 1
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO NCCL_SOCKET_IFNAME set by environment to ens
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO NCCL_SOCKET_IFNAME set to ens
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO Bootstrap : Using ens32:10.200.90.130<0>
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).
a100-st-p4d24xlarge-3:69371:69371 [0] NCCL INFO cudaDriverVersion 11060
NCCL version 2.13.4+cuda11.6
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws
a100-st-p4d24xlarge-3:69371:69878 [0] 28.771366 get_platform_type:1118 NCCL TRACE NET/OFI Read 12 bytes. EC2 platform type is p4d.24xlarge
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/OFI Running on p4d.24xlarge platform, Setting NCCL_TOPO_FILE environment variable to /fsx/users/dberard/scratch-local/bench-fast/aws-ofi-nccl/share/aws-ofi-nccl/xml/p4d-24xl-topo.xml
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/OFI Selected Provider is efa
a100-st-p4d24xlarge-3:69371:69878 [0] 40.029316 ofi_init:1342 NCCL TRACE NET/OFI Provider efa requires registration of local memory buffers
a100-st-p4d24xlarge-3:69371:69878 [0] 40.040884 ofi_init:1349 NCCL TRACE NET/OFI Provider efa requires registration of device buffers
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Using network AWS Libfabric
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NCCL_TOPO_FILE set by environment to /fsx/users/dberard/scratch-local/bench-fast/aws-ofi-nccl/share/aws-ofi-nccl/xml/p4d-24xl-topo.xml
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/AWS Libfabric : GPU Direct RDMA Enabled for HCA 0 'rdmap16s27'
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/AWS Libfabric : GPU Direct RDMA Enabled for HCA 1 'rdmap32s27'
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/AWS Libfabric : GPU Direct RDMA Enabled for HCA 2 'rdmap144s27'
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NET/AWS Libfabric : GPU Direct RDMA Enabled for HCA 3 'rdmap160s27'
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 1 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 2 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 201d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 901d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01c0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU a01d0 / HCA 3 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Setting affinity for GPU 0 to 1f0000,0000001f
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00/08 : 0 7 6 5 4 3 2 1 8 15 14 13 12 11 10 9
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 01/08 : 0 3 10 15 14 13 12 9 8 11 2 7 6 5 4 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 02/08 : 0 7 6 5 12 11 10 9 8 15 14 13 4 3 2 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 03/08 : 0 5 4 7 14 11 10 9 8 13 12 15 6 3 2 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04/08 : 0 7 6 5 4 3 2 1 8 15 14 13 12 11 10 9
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 05/08 : 0 3 10 15 14 13 12 9 8 11 2 7 6 5 4 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 06/08 : 0 7 6 5 12 11 10 9 8 15 14 13 4 3 2 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 07/08 : 0 5 4 7 14 11 10 9 8 13 12 15 6 3 2 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Trees [0] 1/8/-1->0->-1 [1] 1/-1/-1->0->7 [2] 1/-1/-1->0->7 [3] 1/-1/-1->0->7 [4] 1/-1/-1->0->8 [5] 1/-1/-1->0->7 [6] 1/-1/-1->0->7 [7] 1/-1/-1->0->7
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 01 : 0[101c0] -> 3[201d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002d20
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 0 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 05 : 0[101c0] -> 3[201d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002d58
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 1 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 03 : 0[101c0] -> 5[901d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002d90
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 2 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 07 : 0[101c0] -> 5[901d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 3 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002dc8
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 4 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002e00
a100-st-p4d24xlarge-3:69371:69926 [0] 2957.355429 create_nccl_ofi_comp_for_dev:882 NCCL TRACE NET/OFI OFI component #0 is created
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00/0 : 9[101d0] -> 0[101c0] [receive] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 5 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002e38
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04/0 : 9[101d0] -> 0[101c0] [receive] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002e70
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 6 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 02 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 7 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002ea8
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 8 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002ee0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 06 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 9 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002f18
a100-st-p4d24xlarge-3:69371:69926 [0] 3150.481758 alloc_and_reg_flush_buff:2275 NCCL TRACE NET/OFI Registering buffer for flush operations
a100-st-p4d24xlarge-3:69371:69926 [0] 3154.596521 alloc_and_reg_flush_buff:2275 NCCL TRACE NET/OFI Registering buffer for flush operations
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002f50
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 10 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002f88
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 11 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002fc0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 12 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00002ff8
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 13 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003030
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 14 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003068
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 15 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connected all rings
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000030a0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 16 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 01 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000030d8
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 17 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 02 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003110
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 18 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 03 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003148
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 19 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003180
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 20 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 05 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000031b8
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 21 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 06 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000031f0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 22 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 07 : 0[101c0] -> 1[101d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003228
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 23 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 01 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003260
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 24 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 03 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003298
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 25 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 05 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000032d0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 26 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 07 : 0[101c0] -> 7[a01d0] via P2P/IPC/read
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003308
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 27 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 28 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 29 from local rank 5, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 30 from local rank 2, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 31 from local rank 3, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003340
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00/0 : 8[101c0] -> 0[101c0] [receive] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 32 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003420
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04/0 : 8[101c0] -> 0[101c0] [receive] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 33 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003458
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 00/0 : 0[101c0] -> 8[101c0] [send] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 0 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 34 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003490
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Channel 04/0 : 0[101c0] -> 8[101c0] [send] via NET/AWS Libfabric/0/GDRDMA
a100-st-p4d24xlarge-3:69371:69926 [0] 3340.151999 alloc_and_reg_flush_buff:2275 NCCL TRACE NET/OFI Registering buffer for flush operations
a100-st-p4d24xlarge-3:69371:69926 [0] 3341.826148 alloc_and_reg_flush_buff:2275 NCCL TRACE NET/OFI Registering buffer for flush operations
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000034c8
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 35 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003500
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 36 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003538
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 37 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003570
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 38 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000035a8
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 39 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e000035e0
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 40 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003618
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 41 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003650
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy recv connection 42 from local rank 0, transport 0
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connected all trees
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO NCCL_ALGO set by environment to ring
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 8/8/512
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO 8 coll channels, 8 p2p channels, 2 p2p channels per peer
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 43 from local rank 0, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 44 from local rank 1, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 45 from local rank 7, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 46 from local rank 6, transport 2
a100-st-p4d24xlarge-3:69371:69926 [0] NCCL INFO New proxy send connection 47 from local rank 4, transport 2
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 0 -> connection 0x7f3e00003688
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 1 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 2 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO GPU Direct RDMA Enabled for GPU 101c0 / HCA 3 (distance 3 <= 4), read 1
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 2 -> connection 0x7fec90003688
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 4 -> connection 0x7fbee4003768
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO Connection to proxy localRank 6 -> connection 0x7f1294003688
a100-st-p4d24xlarge-3:69371:69878 [0] NCCL INFO comm 0x7f3e04002c30 rank 0 nranks 16 cudaDev 0 busId 101c0 - Init COMPLETE
torchdynamo.eval_frame: [DEBUG] skipping __init__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/contextlib.py
torchdynamo.eval_frame: [DEBUG] skipping __enter__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/contextlib.py
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model [UserDefinedObjectVariable(Model)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR example_inputs [NNModuleVariable(), TupleVariable(), UserDefinedObjectVariable(Model)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_EX 1 [NNModuleVariable(), TupleVariable(), ConstDictVariable()]
torchdynamo.symbolic_convert: [WARNING] Graph break: data dependent operator: aten._local_scalar_dense.default from user code at File "/fsx/users/dberard/scratch-local/bench-fast/benchmark/torchbenchmark/util/framework/huggingface/model_factory.py", line 123, in train
outputs = self.model(**self.example_inputs)
torchdynamo.convert_frame: [INFO] ORIGINAL BYTECODE train /fsx/users/dberard/scratch-local/bench-fast/benchmark/torchbenchmark/util/framework/huggingface/model_factory.py line 122
123 0 LOAD_FAST 0 (self)
2 LOAD_ATTR 0 (model)
4 BUILD_TUPLE 0
6 LOAD_FAST 0 (self)
8 LOAD_ATTR 1 (example_inputs)
10 CALL_FUNCTION_EX 1
12 STORE_FAST 1 (outputs)
124 14 LOAD_FAST 1 (outputs)
16 LOAD_ATTR 2 (loss)
18 STORE_FAST 2 (loss)
125 20 LOAD_FAST 2 (loss)
22 LOAD_METHOD 3 (backward)
24 CALL_METHOD 0
26 POP_TOP
126 28 LOAD_FAST 0 (self)
30 LOAD_ATTR 4 (optimizer)
32 LOAD_METHOD 5 (step)
34 CALL_METHOD 0
36 POP_TOP
38 LOAD_CONST 0 (None)
40 RETURN_VALUE
torchdynamo.convert_frame: [INFO] MODIFIED BYTECODE train /fsx/users/dberard/scratch-local/bench-fast/benchmark/torchbenchmark/util/framework/huggingface/model_factory.py line 122
122 0 LOAD_FAST 0 (self)
2 LOAD_ATTR 0 (model)
4 LOAD_CONST 1 (())
6 LOAD_FAST 0 (self)
8 LOAD_ATTR 1 (example_inputs)
10 CALL_FUNCTION_EX 1
12 LOAD_GLOBAL 6 (__resume_at_12_0)
14 ROT_TWO
16 LOAD_FAST 0 (self)
18 CALL_FUNCTION 2
20 RETURN_VALUE
torchdynamo.convert_frame: [INFO] GUARDS:
-
local 'self' TYPE_MATCH"
{
'guard_types': ['TYPE_MATCH'],
'code': ['___check_type_id(self, 94256889894848)'],
'obj_weakref': <weakref at 0x7f3f1bbcf590; to 'Model' at 0x7f3f1620f8e0>
'guarded_class': <weakref at 0x7f40e12ea360; to 'PostInitProcessor' at 0x55b9e47097c0 (Model)>
}
-
local 'self.model' NN_MODULE"
{
'guard_types': ['ID_MATCH'],
'code': ['___check_obj_id(self.model, 139916925672032)'],
'obj_weakref': <weakref at 0x7f3ef6b17c70; to 'DistributedDataParallel' at 0x7f40f2a71a60>
'guarded_class': <weakref at 0x7f3f1bda6220; to 'ABCMeta' at 0x55b9e32746f0 (DistributedDataParallel)>
}
-
local 'self.example_inputs' DICT_KEYS"
{
'guard_types': ['DICT_KEYS'],
'code': ['___check_type_id(self.example_inputs, 94256774751424)', "set(self.example_inputs.keys()) == {'input_ids', 'labels'}"],
'obj_weakref': None
'guarded_class': <weakref at 0x7f40fc1a00e0; to 'type' at 0x55b9dd93a4c0 (dict)>
}
-
local "self.example_inputs['labels']" TENSOR_MATCH"
{
'guard_types': ['TENSOR_MATCH'],
'code': None,
'obj_weakref': <weakref at 0x7f40f01de450; to 'Tensor' at 0x7f3f1a5de720>
'guarded_class': <weakref at 0x7f3f1c308590; to 'torch._C._TensorMeta' at 0x55b9e2db1e00 (Tensor)>
}
-
local "self.example_inputs['input_ids']" TENSOR_MATCH"
{
'guard_types': ['TENSOR_MATCH'],
'code': None,
'obj_weakref': <weakref at 0x7f40f01c9ef0; to 'Tensor' at 0x7f3f1ad67950>
'guarded_class': <weakref at 0x7f3f1c308590; to 'torch._C._TensorMeta' at 0x55b9e2db1e00 (Tensor)>
}
torchdynamo.eval_frame: [DEBUG] skipping _call_impl /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/modules/module.py
torchdynamo.eval_frame: [DEBUG] skipping forward /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/distributed.py
torchdynamo.eval_frame: [DEBUG] skipping __setattr__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/modules/module.py
torchdynamo.eval_frame: [DEBUG] skipping __instancecheck__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parameter.py
torchdynamo.eval_frame: [DEBUG] skipping notify_join_context /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/algorithms/join.py
torchdynamo.eval_frame: [DEBUG] skipping __getattr__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/modules/module.py
torchdynamo.eval_frame: [DEBUG] skipping _check_sync_bufs_pre_fwd /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/distributed.py
torchdynamo.eval_frame: [DEBUG] skipping will_sync_module_buffers /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/distributed.py
torchdynamo.eval_frame: [DEBUG] skipping _run_ddp_forward /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/distributed.py
torchdynamo.eval_frame: [DEBUG] skipping _to_kwargs /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/utils.py
torchdynamo.eval_frame: [DEBUG] skipping _recursive_to /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/utils.py
torchdynamo.eval_frame: [DEBUG] skipping to_map /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/utils.py
torchdynamo.eval_frame: [DEBUG] skipping _is_namedtuple /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/scatter_gather.py
torchdynamo.eval_frame: [DEBUG] skipping <listcomp> /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/utils.py
torchdynamo.eval_frame: [DEBUG] skipping <listcomp> /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/distributed/utils.py
torchdynamo.eval_frame: [DEBUG] skipping helper /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/contextlib.py
torchdynamo.eval_frame: [DEBUG] skipping __init__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/contextlib.py
torchdynamo.eval_frame: [DEBUG] skipping __enter__ /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/contextlib.py
torchdynamo.eval_frame: [DEBUG] skipping _inside_ddp_forward /data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/parallel/distributed.py
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 12 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR use_cache [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST use_cache [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST return_dict []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 32 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR use_return_dict [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST return_dict [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 88 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR encoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST inputs_embeds [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST return_dict [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('input_ids', 'attention_mask', 'inputs_embeds', 'head_mask', 'output_attentions', 'output_hidden_states', 'return_dict') [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ba030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 892>
908 0 LOAD_FAST 0 (self)
2 LOAD_ATTR 0 (model_parallel)
4 POP_JUMP_IF_FALSE 36
909 6 LOAD_GLOBAL 1 (torch)
8 LOAD_ATTR 2 (cuda)
10 LOAD_METHOD 3 (set_device)
12 LOAD_FAST 0 (self)
14 LOAD_ATTR 4 (first_device)
16 CALL_METHOD 1
18 POP_TOP
910 20 LOAD_FAST 0 (self)
22 LOAD_ATTR 5 (embed_tokens)
24 LOAD_METHOD 6 (to)
26 LOAD_FAST 0 (self)
28 LOAD_ATTR 4 (first_device)
30 CALL_METHOD 1
32 LOAD_FAST 0 (self)
34 STORE_ATTR 5 (embed_tokens)
911 >> 36 LOAD_DEREF 1 (use_cache)
38 LOAD_CONST 0 (None)
40 COMPARE_OP 9 (is not)
42 POP_JUMP_IF_FALSE 48
44 LOAD_DEREF 1 (use_cache)
46 JUMP_FORWARD 6 (to 54)
>> 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (config)
52 LOAD_ATTR 8 (use_cache)
>> 54 STORE_DEREF 1 (use_cache)
912 56 LOAD_DEREF 0 (output_attentions)
58 LOAD_CONST 0 (None)
60 COMPARE_OP 9 (is not)
62 POP_JUMP_IF_FALSE 68
64 LOAD_DEREF 0 (output_attentions)
66 JUMP_FORWARD 6 (to 74)
>> 68 LOAD_FAST 0 (self)
70 LOAD_ATTR 7 (config)
72 LOAD_ATTR 9 (output_attentions)
>> 74 STORE_DEREF 0 (output_attentions)
914 76 LOAD_FAST 11 (output_hidden_states)
78 LOAD_CONST 0 (None)
80 COMPARE_OP 9 (is not)
82 POP_JUMP_IF_FALSE 88
84 LOAD_FAST 11 (output_hidden_states)
86 JUMP_FORWARD 6 (to 94)
>> 88 LOAD_FAST 0 (self)
90 LOAD_ATTR 7 (config)
92 LOAD_ATTR 10 (output_hidden_states)
913 >> 94 STORE_FAST 11 (output_hidden_states)
916 96 LOAD_FAST 12 (return_dict)
98 LOAD_CONST 0 (None)
100 COMPARE_OP 9 (is not)
102 POP_JUMP_IF_FALSE 108
104 LOAD_FAST 12 (return_dict)
106 JUMP_FORWARD 6 (to 114)
>> 108 LOAD_FAST 0 (self)
110 LOAD_ATTR 7 (config)
112 LOAD_ATTR 11 (use_return_dict)
>> 114 STORE_FAST 12 (return_dict)
918 116 LOAD_FAST 1 (input_ids)
118 LOAD_CONST 0 (None)
120 COMPARE_OP 9 (is not)
122 POP_JUMP_IF_FALSE 170
124 LOAD_FAST 5 (inputs_embeds)
126 LOAD_CONST 0 (None)
128 COMPARE_OP 9 (is not)
130 POP_JUMP_IF_FALSE 170
919 132 LOAD_FAST 0 (self)
134 LOAD_ATTR 12 (is_decoder)
136 POP_JUMP_IF_FALSE 142
138 LOAD_CONST 1 ('decoder_')
140 JUMP_FORWARD 2 (to 144)
>> 142 LOAD_CONST 2 ('')
>> 144 STORE_FAST 13 (err_msg_prefix)
920 146 LOAD_GLOBAL 13 (ValueError)
921 148 LOAD_CONST 3 ('You cannot specify both ')
150 LOAD_FAST 13 (err_msg_prefix)
152 FORMAT_VALUE 0
154 LOAD_CONST 4 ('input_ids and ')
156 LOAD_FAST 13 (err_msg_prefix)
158 FORMAT_VALUE 0
160 LOAD_CONST 5 ('inputs_embeds at the same time')
162 BUILD_STRING 5
920 164 CALL_FUNCTION 1
166 RAISE_VARARGS 1
168 JUMP_FORWARD 96 (to 266)
923 >> 170 LOAD_FAST 1 (input_ids)
172 LOAD_CONST 0 (None)
174 COMPARE_OP 9 (is not)
176 POP_JUMP_IF_FALSE 204
924 178 LOAD_FAST 1 (input_ids)
180 LOAD_METHOD 14 (size)
182 CALL_METHOD 0
184 STORE_FAST 14 (input_shape)
925 186 LOAD_FAST 1 (input_ids)
188 LOAD_METHOD 15 (view)
190 LOAD_CONST 6 (-1)
192 LOAD_FAST 14 (input_shape)
194 LOAD_CONST 6 (-1)
196 BINARY_SUBSCR
198 CALL_METHOD 2
200 STORE_FAST 1 (input_ids)
202 JUMP_FORWARD 62 (to 266)
926 >> 204 LOAD_FAST 5 (inputs_embeds)
206 LOAD_CONST 0 (None)
208 COMPARE_OP 9 (is not)
210 POP_JUMP_IF_FALSE 230
927 212 LOAD_FAST 5 (inputs_embeds)
214 LOAD_METHOD 14 (size)
216 CALL_METHOD 0
218 LOAD_CONST 0 (None)
220 LOAD_CONST 6 (-1)
222 BUILD_SLICE 2
224 BINARY_SUBSCR
226 STORE_FAST 14 (input_shape)
228 JUMP_FORWARD 36 (to 266)
929 >> 230 LOAD_FAST 0 (self)
232 LOAD_ATTR 12 (is_decoder)
234 POP_JUMP_IF_FALSE 240
236 LOAD_CONST 1 ('decoder_')
238 JUMP_FORWARD 2 (to 242)
>> 240 LOAD_CONST 2 ('')
>> 242 STORE_FAST 13 (err_msg_prefix)
930 244 LOAD_GLOBAL 13 (ValueError)
246 LOAD_CONST 7 ('You have to specify either ')
248 LOAD_FAST 13 (err_msg_prefix)
250 FORMAT_VALUE 0
252 LOAD_CONST 8 ('input_ids or ')
254 LOAD_FAST 13 (err_msg_prefix)
256 FORMAT_VALUE 0
258 LOAD_CONST 9 ('inputs_embeds')
260 BUILD_STRING 5
262 CALL_FUNCTION 1
264 RAISE_VARARGS 1
932 >> 266 LOAD_FAST 5 (inputs_embeds)
268 LOAD_CONST 0 (None)
270 COMPARE_OP 8 (is)
272 EXTENDED_ARG 1
274 POP_JUMP_IF_FALSE 306
933 276 LOAD_FAST 0 (self)
278 LOAD_ATTR 5 (embed_tokens)
280 LOAD_CONST 0 (None)
282 COMPARE_OP 9 (is not)
284 EXTENDED_ARG 1
286 POP_JUMP_IF_TRUE 296
288 LOAD_GLOBAL 16 (AssertionError)
290 LOAD_CONST 10 ('You have to initialize the model with valid token embeddings')
292 CALL_FUNCTION 1
294 RAISE_VARARGS 1
934 >> 296 LOAD_FAST 0 (self)
298 LOAD_METHOD 5 (embed_tokens)
300 LOAD_FAST 1 (input_ids)
302 CALL_METHOD 1
304 STORE_FAST 5 (inputs_embeds)
936 >> 306 LOAD_FAST 14 (input_shape)
308 UNPACK_SEQUENCE 2
310 STORE_FAST 15 (batch_size)
312 STORE_FAST 16 (seq_length)
939 314 LOAD_FAST 8 (past_key_values)
316 LOAD_CONST 0 (None)
318 COMPARE_OP 9 (is not)
320 EXTENDED_ARG 1
322 POP_JUMP_IF_FALSE 346
324 LOAD_FAST 8 (past_key_values)
326 LOAD_CONST 11 (0)
328 BINARY_SUBSCR
330 LOAD_CONST 11 (0)
332 BINARY_SUBSCR
334 LOAD_ATTR 17 (shape)
336 LOAD_CONST 12 (2)
338 BINARY_SUBSCR
340 LOAD_FAST 16 (seq_length)
342 BINARY_ADD
344 JUMP_FORWARD 2 (to 348)
>> 346 LOAD_FAST 16 (seq_length)
>> 348 STORE_FAST 17 (mask_seq_length)
941 350 LOAD_DEREF 1 (use_cache)
352 LOAD_CONST 13 (True)
354 COMPARE_OP 8 (is)
356 EXTENDED_ARG 1
358 POP_JUMP_IF_FALSE 384
942 360 LOAD_FAST 0 (self)
362 LOAD_ATTR 12 (is_decoder)
364 EXTENDED_ARG 1
366 POP_JUMP_IF_TRUE 384
368 LOAD_GLOBAL 16 (AssertionError)
370 LOAD_CONST 14 ('`use_cache` can only be set to `True` if ')
372 LOAD_FAST 0 (self)
374 FORMAT_VALUE 0
376 LOAD_CONST 15 (' is used as a decoder')
378 BUILD_STRING 3
380 CALL_FUNCTION 1
382 RAISE_VARARGS 1
944 >> 384 LOAD_FAST 2 (attention_mask)
386 LOAD_CONST 0 (None)
388 COMPARE_OP 8 (is)
390 EXTENDED_ARG 1
392 POP_JUMP_IF_FALSE 414
945 394 LOAD_GLOBAL 1 (torch)
396 LOAD_METHOD 18 (ones)
398 LOAD_FAST 15 (batch_size)
400 LOAD_FAST 17 (mask_seq_length)
402 CALL_METHOD 2
404 LOAD_METHOD 6 (to)
406 LOAD_FAST 5 (inputs_embeds)
408 LOAD_ATTR 19 (device)
410 CALL_METHOD 1
412 STORE_FAST 2 (attention_mask)
946 >> 414 LOAD_FAST 0 (self)
416 LOAD_ATTR 12 (is_decoder)
418 EXTENDED_ARG 1
420 POP_JUMP_IF_FALSE 474
422 LOAD_FAST 4 (encoder_attention_mask)
424 LOAD_CONST 0 (None)
426 COMPARE_OP 8 (is)
428 EXTENDED_ARG 1
430 POP_JUMP_IF_FALSE 474
432 LOAD_FAST 3 (encoder_hidden_states)
434 LOAD_CONST 0 (None)
436 COMPARE_OP 9 (is not)
438 EXTENDED_ARG 1
440 POP_JUMP_IF_FALSE 474
947 442 LOAD_FAST 3 (encoder_hidden_states)
444 LOAD_ATTR 17 (shape)
446 LOAD_CONST 16 (1)
448 BINARY_SUBSCR
450 STORE_FAST 18 (encoder_seq_length)
948 452 LOAD_GLOBAL 1 (torch)
454 LOAD_ATTR 18 (ones)
949 456 LOAD_FAST 15 (batch_size)
458 LOAD_FAST 18 (encoder_seq_length)
460 LOAD_FAST 5 (inputs_embeds)
462 LOAD_ATTR 19 (device)
464 LOAD_GLOBAL 1 (torch)
466 LOAD_ATTR 20 (long)
948 468 LOAD_CONST 17 (('device', 'dtype'))
470 CALL_FUNCTION_KW 4
472 STORE_FAST 4 (encoder_attention_mask)
953 >> 474 LOAD_FAST 8 (past_key_values)
476 LOAD_CONST 0 (None)
478 COMPARE_OP 8 (is)
480 EXTENDED_ARG 1
482 POP_JUMP_IF_FALSE 500
954 484 LOAD_CONST 0 (None)
486 BUILD_LIST 1
488 LOAD_GLOBAL 21 (len)
490 LOAD_FAST 0 (self)
492 LOAD_ATTR 22 (block)
494 CALL_FUNCTION 1
496 BINARY_MULTIPLY
498 STORE_FAST 8 (past_key_values)
958 >> 500 LOAD_FAST 0 (self)
502 LOAD_METHOD 23 (get_extended_attention_mask)
504 LOAD_FAST 2 (attention_mask)
506 LOAD_FAST 14 (input_shape)
508 CALL_METHOD 2
510 STORE_FAST 19 (extended_attention_mask)
962 512 LOAD_FAST 0 (self)
514 LOAD_ATTR 12 (is_decoder)
516 EXTENDED_ARG 2
518 POP_JUMP_IF_FALSE 590
520 LOAD_FAST 3 (encoder_hidden_states)
522 LOAD_CONST 0 (None)
524 COMPARE_OP 9 (is not)
526 EXTENDED_ARG 2
528 POP_JUMP_IF_FALSE 590
963 530 LOAD_FAST 3 (encoder_hidden_states)
532 LOAD_METHOD 14 (size)
534 CALL_METHOD 0
536 UNPACK_SEQUENCE 3
538 STORE_FAST 20 (encoder_batch_size)
540 STORE_FAST 21 (encoder_sequence_length)
542 STORE_FAST 22 (_)
964 544 LOAD_FAST 20 (encoder_batch_size)
546 LOAD_FAST 21 (encoder_sequence_length)
548 BUILD_TUPLE 2
550 STORE_FAST 23 (encoder_hidden_shape)
965 552 LOAD_FAST 4 (encoder_attention_mask)
554 LOAD_CONST 0 (None)
556 COMPARE_OP 8 (is)
558 EXTENDED_ARG 2
560 POP_JUMP_IF_FALSE 578
966 562 LOAD_GLOBAL 1 (torch)
564 LOAD_ATTR 18 (ones)
566 LOAD_FAST 23 (encoder_hidden_shape)
568 LOAD_FAST 5 (inputs_embeds)
570 LOAD_ATTR 19 (device)
572 LOAD_CONST 18 (('device',))
574 CALL_FUNCTION_KW 2
576 STORE_FAST 4 (encoder_attention_mask)
967 >> 578 LOAD_FAST 0 (self)
580 LOAD_METHOD 24 (invert_attention_mask)
582 LOAD_FAST 4 (encoder_attention_mask)
584 CALL_METHOD 1
586 STORE_FAST 24 (encoder_extended_attention_mask)
588 JUMP_FORWARD 4 (to 594)
969 >> 590 LOAD_CONST 0 (None)
592 STORE_FAST 24 (encoder_extended_attention_mask)
972 >> 594 LOAD_FAST 0 (self)
596 LOAD_METHOD 25 (get_head_mask)
598 LOAD_FAST 6 (head_mask)
600 LOAD_FAST 0 (self)
602 LOAD_ATTR 7 (config)
604 LOAD_ATTR 26 (num_layers)
606 CALL_METHOD 2
608 STORE_FAST 6 (head_mask)
973 610 LOAD_FAST 0 (self)
612 LOAD_METHOD 25 (get_head_mask)
614 LOAD_FAST 7 (cross_attn_head_mask)
616 LOAD_FAST 0 (self)
618 LOAD_ATTR 7 (config)
620 LOAD_ATTR 26 (num_layers)
622 CALL_METHOD 2
624 STORE_FAST 7 (cross_attn_head_mask)
974 626 LOAD_DEREF 1 (use_cache)
628 EXTENDED_ARG 2
630 POP_JUMP_IF_FALSE 636
632 LOAD_CONST 19 (())
634 JUMP_FORWARD 2 (to 638)
>> 636 LOAD_CONST 0 (None)
>> 638 STORE_FAST 25 (present_key_value_states)
975 640 LOAD_FAST 11 (output_hidden_states)
642 EXTENDED_ARG 2
644 POP_JUMP_IF_FALSE 650
646 LOAD_CONST 19 (())
648 JUMP_FORWARD 2 (to 652)
>> 650 LOAD_CONST 0 (None)
>> 652 STORE_FAST 26 (all_hidden_states)
976 654 LOAD_DEREF 0 (output_attentions)
656 EXTENDED_ARG 2
658 POP_JUMP_IF_FALSE 664
660 LOAD_CONST 19 (())
662 JUMP_FORWARD 2 (to 666)
>> 664 LOAD_CONST 0 (None)
>> 666 STORE_FAST 27 (all_attentions)
977 668 LOAD_DEREF 0 (output_attentions)
670 EXTENDED_ARG 2
672 POP_JUMP_IF_FALSE 686
674 LOAD_FAST 0 (self)
676 LOAD_ATTR 12 (is_decoder)
678 EXTENDED_ARG 2
680 POP_JUMP_IF_FALSE 686
682 LOAD_CONST 19 (())
684 JUMP_FORWARD 2 (to 688)
>> 686 LOAD_CONST 0 (None)
>> 688 STORE_FAST 28 (all_cross_attentions)
978 690 LOAD_CONST 0 (None)
692 STORE_FAST 29 (position_bias)
979 694 LOAD_CONST 0 (None)
696 STORE_FAST 30 (encoder_decoder_position_bias)
981 698 LOAD_FAST 0 (self)
700 LOAD_METHOD 27 (dropout)
702 LOAD_FAST 5 (inputs_embeds)
704 CALL_METHOD 1
706 STORE_FAST 31 (hidden_states)
983 708 LOAD_GLOBAL 28 (enumerate)
710 LOAD_GLOBAL 29 (zip)
712 LOAD_FAST 0 (self)
714 LOAD_ATTR 22 (block)
716 LOAD_FAST 8 (past_key_values)
718 CALL_FUNCTION 2
720 CALL_FUNCTION 1
722 GET_ITER
>> 724 EXTENDED_ARG 2
726 FOR_ITER 576 (to 1304)
728 UNPACK_SEQUENCE 2
730 STORE_FAST 32 (i)
732 UNPACK_SEQUENCE 2
734 STORE_FAST 33 (layer_module)
736 STORE_FAST 34 (past_key_value)
984 738 LOAD_FAST 6 (head_mask)
740 LOAD_FAST 32 (i)
742 BINARY_SUBSCR
744 STORE_FAST 35 (layer_head_mask)
985 746 LOAD_FAST 7 (cross_attn_head_mask)
748 LOAD_FAST 32 (i)
750 BINARY_SUBSCR
752 STORE_FAST 36 (cross_attn_layer_head_mask)
987 754 LOAD_FAST 0 (self)
756 LOAD_ATTR 0 (model_parallel)
758 EXTENDED_ARG 3
760 POP_JUMP_IF_FALSE 930
988 762 LOAD_GLOBAL 1 (torch)
764 LOAD_ATTR 2 (cuda)
766 LOAD_METHOD 3 (set_device)
768 LOAD_FAST 31 (hidden_states)
770 LOAD_ATTR 19 (device)
772 CALL_METHOD 1
774 POP_TOP
990 776 LOAD_FAST 2 (attention_mask)
778 LOAD_CONST 0 (None)
780 COMPARE_OP 9 (is not)
782 EXTENDED_ARG 3
784 POP_JUMP_IF_FALSE 798
991 786 LOAD_FAST 2 (attention_mask)
788 LOAD_METHOD 6 (to)
790 LOAD_FAST 31 (hidden_states)
792 LOAD_ATTR 19 (device)
794 CALL_METHOD 1
796 STORE_FAST 2 (attention_mask)
992 >> 798 LOAD_FAST 29 (position_bias)
800 LOAD_CONST 0 (None)
802 COMPARE_OP 9 (is not)
804 EXTENDED_ARG 3
806 POP_JUMP_IF_FALSE 820
993 808 LOAD_FAST 29 (position_bias)
810 LOAD_METHOD 6 (to)
812 LOAD_FAST 31 (hidden_states)
814 LOAD_ATTR 19 (device)
816 CALL_METHOD 1
818 STORE_FAST 29 (position_bias)
994 >> 820 LOAD_FAST 3 (encoder_hidden_states)
822 LOAD_CONST 0 (None)
824 COMPARE_OP 9 (is not)
826 EXTENDED_ARG 3
828 POP_JUMP_IF_FALSE 842
995 830 LOAD_FAST 3 (encoder_hidden_states)
832 LOAD_METHOD 6 (to)
834 LOAD_FAST 31 (hidden_states)
836 LOAD_ATTR 19 (device)
838 CALL_METHOD 1
840 STORE_FAST 3 (encoder_hidden_states)
996 >> 842 LOAD_FAST 24 (encoder_extended_attention_mask)
844 LOAD_CONST 0 (None)
846 COMPARE_OP 9 (is not)
848 EXTENDED_ARG 3
850 POP_JUMP_IF_FALSE 864
997 852 LOAD_FAST 24 (encoder_extended_attention_mask)
854 LOAD_METHOD 6 (to)
856 LOAD_FAST 31 (hidden_states)
858 LOAD_ATTR 19 (device)
860 CALL_METHOD 1
862 STORE_FAST 24 (encoder_extended_attention_mask)
998 >> 864 LOAD_FAST 30 (encoder_decoder_position_bias)
866 LOAD_CONST 0 (None)
868 COMPARE_OP 9 (is not)
870 EXTENDED_ARG 3
872 POP_JUMP_IF_FALSE 886
999 874 LOAD_FAST 30 (encoder_decoder_position_bias)
876 LOAD_METHOD 6 (to)
878 LOAD_FAST 31 (hidden_states)
880 LOAD_ATTR 19 (device)
882 CALL_METHOD 1
884 STORE_FAST 30 (encoder_decoder_position_bias)
1000 >> 886 LOAD_FAST 35 (layer_head_mask)
888 LOAD_CONST 0 (None)
890 COMPARE_OP 9 (is not)
892 EXTENDED_ARG 3
894 POP_JUMP_IF_FALSE 908
1001 896 LOAD_FAST 35 (layer_head_mask)
898 LOAD_METHOD 6 (to)
900 LOAD_FAST 31 (hidden_states)
902 LOAD_ATTR 19 (device)
904 CALL_METHOD 1
906 STORE_FAST 35 (layer_head_mask)
1002 >> 908 LOAD_FAST 36 (cross_attn_layer_head_mask)
910 LOAD_CONST 0 (None)
912 COMPARE_OP 9 (is not)
914 EXTENDED_ARG 3
916 POP_JUMP_IF_FALSE 930
1003 918 LOAD_FAST 36 (cross_attn_layer_head_mask)
920 LOAD_METHOD 6 (to)
922 LOAD_FAST 31 (hidden_states)
924 LOAD_ATTR 19 (device)
926 CALL_METHOD 1
928 STORE_FAST 36 (cross_attn_layer_head_mask)
1004 >> 930 LOAD_FAST 11 (output_hidden_states)
932 EXTENDED_ARG 3
934 POP_JUMP_IF_FALSE 946
1005 936 LOAD_FAST 26 (all_hidden_states)
938 LOAD_FAST 31 (hidden_states)
940 BUILD_TUPLE 1
942 BINARY_ADD
944 STORE_FAST 26 (all_hidden_states)
1007 >> 946 LOAD_FAST 0 (self)
948 LOAD_ATTR 30 (gradient_checkpointing)
950 EXTENDED_ARG 4
952 POP_JUMP_IF_FALSE 1028
954 LOAD_FAST 0 (self)
956 LOAD_ATTR 31 (training)
958 EXTENDED_ARG 4
960 POP_JUMP_IF_FALSE 1028
1008 962 LOAD_DEREF 1 (use_cache)
964 EXTENDED_ARG 3
966 POP_JUMP_IF_FALSE 982
1009 968 LOAD_GLOBAL 32 (logger)
970 LOAD_METHOD 33 (warning)
1010 972 LOAD_CONST 20 ('`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...')
1009 974 CALL_METHOD 1
976 POP_TOP
1012 978 LOAD_CONST 21 (False)
980 STORE_DEREF 1 (use_cache)
1014 >> 982 LOAD_CLOSURE 0 (output_attentions)
984 LOAD_CLOSURE 1 (use_cache)
986 BUILD_TUPLE 2
988 LOAD_CONST 22 (<code object create_custom_forward at 0x7f3f100b4ea0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 1014>)
990 LOAD_CONST 23 ('T5Stack.forward.<locals>.create_custom_forward')
992 MAKE_FUNCTION 8 (closure)
994 STORE_FAST 37 (create_custom_forward)
1020 996 LOAD_GLOBAL 34 (checkpoint)
1021 998 LOAD_FAST 37 (create_custom_forward)
1000 LOAD_FAST 33 (layer_module)
1002 CALL_FUNCTION 1
1022 1004 LOAD_FAST 31 (hidden_states)
1023 1006 LOAD_FAST 19 (extended_attention_mask)
1024 1008 LOAD_FAST 29 (position_bias)
1025 1010 LOAD_FAST 3 (encoder_hidden_states)
1026 1012 LOAD_FAST 24 (encoder_extended_attention_mask)
1027 1014 LOAD_FAST 30 (encoder_decoder_position_bias)
1028 1016 LOAD_FAST 35 (layer_head_mask)
1029 1018 LOAD_FAST 36 (cross_attn_layer_head_mask)
1030 1020 LOAD_CONST 0 (None)
1020 1022 CALL_FUNCTION 10
1024 STORE_FAST 38 (layer_outputs)
1026 JUMP_FORWARD 30 (to 1058)
1033 >> 1028 LOAD_FAST 33 (layer_module)
1034 1030 LOAD_FAST 31 (hidden_states)
1035 1032 LOAD_FAST 19 (extended_attention_mask)
1036 1034 LOAD_FAST 29 (position_bias)
1037 1036 LOAD_FAST 3 (encoder_hidden_states)
1038 1038 LOAD_FAST 24 (encoder_extended_attention_mask)
1039 1040 LOAD_FAST 30 (encoder_decoder_position_bias)
1040 1042 LOAD_FAST 35 (layer_head_mask)
1041 1044 LOAD_FAST 36 (cross_attn_layer_head_mask)
1042 1046 LOAD_FAST 34 (past_key_value)
1043 1048 LOAD_DEREF 1 (use_cache)
1044 1050 LOAD_DEREF 0 (output_attentions)
1033 1052 LOAD_CONST 24 (('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
1054 CALL_FUNCTION_KW 11
1056 STORE_FAST 38 (layer_outputs)
1049 >> 1058 LOAD_DEREF 1 (use_cache)
1060 LOAD_CONST 21 (False)
1062 COMPARE_OP 8 (is)
1064 EXTENDED_ARG 4
1066 POP_JUMP_IF_FALSE 1096
1050 1068 LOAD_FAST 38 (layer_outputs)
1070 LOAD_CONST 0 (None)
1072 LOAD_CONST 16 (1)
1074 BUILD_SLICE 2
1076 BINARY_SUBSCR
1078 LOAD_CONST 25 ((None,))
1080 BINARY_ADD
1082 LOAD_FAST 38 (layer_outputs)
1084 LOAD_CONST 16 (1)
1086 LOAD_CONST 0 (None)
1088 BUILD_SLICE 2
1090 BINARY_SUBSCR
1092 BINARY_ADD
1094 STORE_FAST 38 (layer_outputs)
1052 >> 1096 LOAD_FAST 38 (layer_outputs)
1098 LOAD_CONST 0 (None)
1100 LOAD_CONST 12 (2)
1102 BUILD_SLICE 2
1104 BINARY_SUBSCR
1106 UNPACK_SEQUENCE 2
1108 STORE_FAST 31 (hidden_states)
1110 STORE_FAST 39 (present_key_value_state)
1057 1112 LOAD_FAST 38 (layer_outputs)
1114 LOAD_CONST 12 (2)
1116 BINARY_SUBSCR
1118 STORE_FAST 29 (position_bias)
1058 1120 LOAD_FAST 0 (self)
1122 LOAD_ATTR 12 (is_decoder)
1124 EXTENDED_ARG 4
1126 POP_JUMP_IF_FALSE 1156
1128 LOAD_FAST 3 (encoder_hidden_states)
1130 LOAD_CONST 0 (None)
1132 COMPARE_OP 9 (is not)
1134 EXTENDED_ARG 4
1136 POP_JUMP_IF_FALSE 1156
1059 1138 LOAD_FAST 38 (layer_outputs)
1140 LOAD_DEREF 0 (output_attentions)
1142 EXTENDED_ARG 4
1144 POP_JUMP_IF_FALSE 1150
1146 LOAD_CONST 26 (4)
1148 JUMP_FORWARD 2 (to 1152)
>> 1150 LOAD_CONST 27 (3)
>> 1152 BINARY_SUBSCR
1154 STORE_FAST 30 (encoder_decoder_position_bias)
1061 >> 1156 LOAD_DEREF 1 (use_cache)
1158 EXTENDED_ARG 4
1160 POP_JUMP_IF_FALSE 1172
1062 1162 LOAD_FAST 25 (present_key_value_states)
1164 LOAD_FAST 39 (present_key_value_state)
1166 BUILD_TUPLE 1
1168 BINARY_ADD
1170 STORE_FAST 25 (present_key_value_states)
1064 >> 1172 LOAD_DEREF 0 (output_attentions)
1174 EXTENDED_ARG 4
1176 POP_JUMP_IF_FALSE 1214
1065 1178 LOAD_FAST 27 (all_attentions)
1180 LOAD_FAST 38 (layer_outputs)
1182 LOAD_CONST 27 (3)
1184 BINARY_SUBSCR
1186 BUILD_TUPLE 1
1188 BINARY_ADD
1190 STORE_FAST 27 (all_attentions)
1066 1192 LOAD_FAST 0 (self)
1194 LOAD_ATTR 12 (is_decoder)
1196 EXTENDED_ARG 4
1198 POP_JUMP_IF_FALSE 1214
1067 1200 LOAD_FAST 28 (all_cross_attentions)
1202 LOAD_FAST 38 (layer_outputs)
1204 LOAD_CONST 28 (5)
1206 BINARY_SUBSCR
1208 BUILD_TUPLE 1
1210 BINARY_ADD
1212 STORE_FAST 28 (all_cross_attentions)
1070 >> 1214 LOAD_FAST 0 (self)
1216 LOAD_ATTR 0 (model_parallel)
1218 EXTENDED_ARG 2
1220 POP_JUMP_IF_FALSE 724
1071 1222 LOAD_FAST 0 (self)
1224 LOAD_ATTR 35 (device_map)
1226 LOAD_METHOD 36 (items)
1228 CALL_METHOD 0
1230 GET_ITER
>> 1232 FOR_ITER 66 (to 1300)
1234 UNPACK_SEQUENCE 2
1236 STORE_FAST 40 (k)
1238 STORE_FAST 41 (v)
1072 1240 LOAD_FAST 32 (i)
1242 LOAD_FAST 41 (v)
1244 LOAD_CONST 6 (-1)
1246 BINARY_SUBSCR
1248 COMPARE_OP 2 (==)
1250 EXTENDED_ARG 4
1252 POP_JUMP_IF_FALSE 1232
1254 LOAD_CONST 29 ('cuda:')
1256 LOAD_GLOBAL 37 (str)
1258 LOAD_FAST 40 (k)
1260 CALL_FUNCTION 1
1262 BINARY_ADD
1264 LOAD_FAST 0 (self)
1266 LOAD_ATTR 38 (last_device)
1268 COMPARE_OP 3 (!=)
1270 EXTENDED_ARG 4
1272 POP_JUMP_IF_FALSE 1232
1073 1274 LOAD_FAST 31 (hidden_states)
1276 LOAD_METHOD 6 (to)
1278 LOAD_CONST 29 ('cuda:')
1280 LOAD_GLOBAL 37 (str)
1282 LOAD_FAST 40 (k)
1284 LOAD_CONST 16 (1)
1286 BINARY_ADD
1288 CALL_FUNCTION 1
1290 BINARY_ADD
1292 CALL_METHOD 1
1294 STORE_FAST 31 (hidden_states)
1296 EXTENDED_ARG 4
1298 JUMP_ABSOLUTE 1232
>> 1300 EXTENDED_ARG 2
1302 JUMP_ABSOLUTE 724
1075 >> 1304 LOAD_FAST 0 (self)
1306 LOAD_METHOD 39 (final_layer_norm)
1308 LOAD_FAST 31 (hidden_states)
1310 CALL_METHOD 1
1312 STORE_FAST 31 (hidden_states)
1076 1314 LOAD_FAST 0 (self)
1316 LOAD_METHOD 27 (dropout)
1318 LOAD_FAST 31 (hidden_states)
1320 CALL_METHOD 1
1322 STORE_FAST 31 (hidden_states)
1079 1324 LOAD_FAST 11 (output_hidden_states)
1326 EXTENDED_ARG 5
1328 POP_JUMP_IF_FALSE 1340
1080 1330 LOAD_FAST 26 (all_hidden_states)
1332 LOAD_FAST 31 (hidden_states)
1334 BUILD_TUPLE 1
1336 BINARY_ADD
1338 STORE_FAST 26 (all_hidden_states)
1082 >> 1340 LOAD_FAST 12 (return_dict)
1342 EXTENDED_ARG 5
1344 POP_JUMP_IF_TRUE 1374
1083 1346 LOAD_GLOBAL 40 (tuple)
1348 LOAD_CONST 30 (<code object <genexpr> at 0x7f3f100b4f50, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 1083>)
1350 LOAD_CONST 31 ('T5Stack.forward.<locals>.<genexpr>')
1352 MAKE_FUNCTION 0
1086 1354 LOAD_FAST 31 (hidden_states)
1087 1356 LOAD_FAST 25 (present_key_value_states)
1088 1358 LOAD_FAST 26 (all_hidden_states)
1089 1360 LOAD_FAST 27 (all_attentions)
1090 1362 LOAD_FAST 28 (all_cross_attentions)
1085 1364 BUILD_TUPLE 5
1083 1366 GET_ITER
1368 CALL_FUNCTION 1
1370 CALL_FUNCTION 1
1372 RETURN_VALUE
1094 >> 1374 LOAD_GLOBAL 41 (BaseModelOutputWithPastAndCrossAttentions)
1095 1376 LOAD_FAST 31 (hidden_states)
1096 1378 LOAD_FAST 25 (present_key_value_states)
1097 1380 LOAD_FAST 26 (all_hidden_states)
1098 1382 LOAD_FAST 27 (all_attentions)
1099 1384 LOAD_FAST 28 (all_cross_attentions)
1094 1386 LOAD_CONST 32 (('last_hidden_state', 'past_key_values', 'hidden_states', 'attentions', 'cross_attentions'))
1388 CALL_FUNCTION_KW 5
1390 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 36 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 48 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR use_cache [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF use_cache [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 68 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR output_attentions [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF output_attentions [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 88 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR output_hidden_states [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST output_hidden_states [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST return_dict []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(bool), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 108 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST return_dict []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 114 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST return_dict [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 170 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST inputs_embeds []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 170 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 204 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR size [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), size)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST input_shape [SizeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_shape [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), SizeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), SizeVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST input_ids [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 266 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST inputs_embeds []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 306 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR embed_tokens [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 296 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR embed_tokens [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST inputs_embeds [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [SizeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_values []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 346 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST mask_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 384 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 414 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR ones [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST batch_size [TorchVariable(<built-in method ones of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST mask_seq_length [TorchVariable(<built-in method ones of type object at 0x7f40b0102e20>), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method ones of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST inputs_embeds [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR device [GetAttrVariable(TensorVariable(), to), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), TorchVariable(cuda:0)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 474 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_values []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 500 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_LIST 1 [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL len [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListVariable(), BuiltinVariable(len)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR block [ListVariable(), BuiltinVariable(len), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [ListVariable(), BuiltinVariable(len), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_values [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR get_extended_attention_mask [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [UserMethodVariable(<function ModuleUtilsMixin.get_extended_attention_mask at 0x7f3ef770e280>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_shape [UserMethodVariable(<function ModuleUtilsMixin.get_extended_attention_mask at 0x7f3ef770e280>, NNModuleVariable()), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [UserMethodVariable(<function ModuleUtilsMixin.get_extended_attention_mask at 0x7f3ef770e280>, NNModuleVariable()), TensorVariable(), SizeVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object get_extended_attention_mask at 0x7f3f10087be0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 794>
809 0 LOAD_FAST 1 (attention_mask)
2 LOAD_METHOD 0 (dim)
4 CALL_METHOD 0
6 LOAD_CONST 1 (2)
8 COMPARE_OP 2 (==)
10 POP_JUMP_IF_FALSE 20
12 LOAD_FAST 0 (self)
14 LOAD_ATTR 1 (config)
16 LOAD_ATTR 2 (is_decoder)
18 POP_JUMP_IF_TRUE 40
811 >> 20 LOAD_FAST 3 (device)
22 LOAD_CONST 2 (None)
24 COMPARE_OP 9 (is not)
26 POP_JUMP_IF_FALSE 40
812 28 LOAD_GLOBAL 3 (warnings)
30 LOAD_METHOD 4 (warn)
813 32 LOAD_CONST 3 ('The `device` argument is deprecated and will be removed in v5 of Transformers.')
34 LOAD_GLOBAL 5 (FutureWarning)
812 36 CALL_METHOD 2
38 POP_TOP
817 >> 40 LOAD_FAST 1 (attention_mask)
42 LOAD_METHOD 0 (dim)
44 CALL_METHOD 0
46 LOAD_CONST 4 (3)
48 COMPARE_OP 2 (==)
50 POP_JUMP_IF_FALSE 82
818 52 LOAD_FAST 1 (attention_mask)
54 LOAD_CONST 2 (None)
56 LOAD_CONST 2 (None)
58 BUILD_SLICE 2
60 LOAD_CONST 2 (None)
62 LOAD_CONST 2 (None)
64 LOAD_CONST 2 (None)
66 BUILD_SLICE 2
68 LOAD_CONST 2 (None)
70 LOAD_CONST 2 (None)
72 BUILD_SLICE 2
74 BUILD_TUPLE 4
76 BINARY_SUBSCR
78 STORE_FAST 4 (extended_attention_mask)
80 JUMP_FORWARD 86 (to 168)
819 >> 82 LOAD_FAST 1 (attention_mask)
84 LOAD_METHOD 0 (dim)
86 CALL_METHOD 0
88 LOAD_CONST 1 (2)
90 COMPARE_OP 2 (==)
92 POP_JUMP_IF_FALSE 144
823 94 LOAD_FAST 0 (self)
96 LOAD_ATTR 1 (config)
98 LOAD_ATTR 2 (is_decoder)
100 POP_JUMP_IF_FALSE 118
824 102 LOAD_GLOBAL 6 (ModuleUtilsMixin)
104 LOAD_METHOD 7 (create_extended_attention_mask_for_decoder)
825 106 LOAD_FAST 2 (input_shape)
108 LOAD_FAST 1 (attention_mask)
110 LOAD_FAST 3 (device)
824 112 CALL_METHOD 3
114 STORE_FAST 4 (extended_attention_mask)
116 JUMP_ABSOLUTE 168
828 >> 118 LOAD_FAST 1 (attention_mask)
120 LOAD_CONST 2 (None)
122 LOAD_CONST 2 (None)
124 BUILD_SLICE 2
126 LOAD_CONST 2 (None)
128 LOAD_CONST 2 (None)
130 LOAD_CONST 2 (None)
132 LOAD_CONST 2 (None)
134 BUILD_SLICE 2
136 BUILD_TUPLE 4
138 BINARY_SUBSCR
140 STORE_FAST 4 (extended_attention_mask)
142 JUMP_FORWARD 24 (to 168)
830 >> 144 LOAD_GLOBAL 8 (ValueError)
831 146 LOAD_CONST 5 ('Wrong shape for input_ids (shape ')
148 LOAD_FAST 2 (input_shape)
150 FORMAT_VALUE 0
152 LOAD_CONST 6 (') or attention_mask (shape ')
154 LOAD_FAST 1 (attention_mask)
156 LOAD_ATTR 9 (shape)
158 FORMAT_VALUE 0
160 LOAD_CONST 7 (')')
162 BUILD_STRING 5
830 164 CALL_FUNCTION 1
166 RAISE_VARARGS 1
839 >> 168 LOAD_FAST 4 (extended_attention_mask)
170 LOAD_ATTR 10 (to)
172 LOAD_FAST 0 (self)
174 LOAD_ATTR 11 (dtype)
176 LOAD_CONST 8 (('dtype',))
178 CALL_FUNCTION_KW 1
180 STORE_FAST 4 (extended_attention_mask)
840 182 LOAD_CONST 9 (1.0)
184 LOAD_FAST 4 (extended_attention_mask)
186 BINARY_SUBTRACT
188 LOAD_CONST 10 (-10000.0)
190 BINARY_MULTIPLY
192 STORE_FAST 4 (extended_attention_mask)
841 194 LOAD_FAST 4 (extended_attention_mask)
196 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dim [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), dim)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 20 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 40 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST device []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 40 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dim [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), dim)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dim [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), dim)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 144 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 118 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), SliceVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), SliceVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), SliceVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), SliceVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 4 [TensorVariable(), SliceVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST extended_attention_mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 168 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [GetAttrVariable(TensorVariable(), to), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object dtype at 0x7f3f100879d0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 727>
732 0 LOAD_GLOBAL 0 (get_parameter_dtype)
2 LOAD_FAST 0 (self)
4 CALL_FUNCTION 1
6 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL get_parameter_dtype []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [UserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object get_parameter_dtype at 0x7f3f1005d190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 153>
157 0 LOAD_CONST 1 (None)
2 STORE_FAST 1 (last_dtype)
158 4 LOAD_FAST 0 (parameter)
6 LOAD_METHOD 0 (parameters)
8 CALL_METHOD 0
10 GET_ITER
>> 12 FOR_ITER 28 (to 42)
14 STORE_FAST 2 (t)
159 16 LOAD_FAST 2 (t)
18 LOAD_ATTR 1 (dtype)
20 STORE_FAST 1 (last_dtype)
160 22 LOAD_FAST 2 (t)
24 LOAD_METHOD 2 (is_floating_point)
26 CALL_METHOD 0
28 POP_JUMP_IF_FALSE 12
161 30 LOAD_FAST 2 (t)
32 LOAD_ATTR 1 (dtype)
34 ROT_TWO
36 POP_TOP
38 RETURN_VALUE
40 JUMP_ABSOLUTE 12
163 >> 42 LOAD_FAST 1 (last_dtype)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 54
165 50 LOAD_FAST 1 (last_dtype)
52 RETURN_VALUE
169 >> 54 LOAD_GLOBAL 3 (nn)
56 LOAD_ATTR 4 (Module)
58 LOAD_GLOBAL 5 (List)
60 LOAD_GLOBAL 6 (Tuple)
62 LOAD_GLOBAL 7 (str)
64 LOAD_GLOBAL 8 (Tensor)
66 BUILD_TUPLE 2
68 BINARY_SUBSCR
70 BINARY_SUBSCR
72 LOAD_CONST 2 (('module', 'return'))
74 BUILD_CONST_KEY_MAP 2
76 LOAD_CONST 3 (<code object find_tensor_attributes at 0x7f3f1005d030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 169>)
78 LOAD_CONST 4 ('get_parameter_dtype.<locals>.find_tensor_attributes')
80 MAKE_FUNCTION 4 (annotations)
82 STORE_FAST 3 (find_tensor_attributes)
173 84 LOAD_FAST 0 (parameter)
86 LOAD_ATTR 9 (_named_members)
88 LOAD_FAST 3 (find_tensor_attributes)
90 LOAD_CONST 5 (('get_members_fn',))
92 CALL_FUNCTION_KW 1
94 STORE_FAST 4 (gen)
174 96 LOAD_CONST 1 (None)
98 STORE_FAST 5 (last_tuple)
175 100 LOAD_FAST 4 (gen)
102 GET_ITER
>> 104 FOR_ITER 34 (to 140)
106 STORE_FAST 6 (tuple)
176 108 LOAD_FAST 6 (tuple)
110 STORE_FAST 5 (last_tuple)
177 112 LOAD_FAST 6 (tuple)
114 LOAD_CONST 6 (1)
116 BINARY_SUBSCR
118 LOAD_METHOD 2 (is_floating_point)
120 CALL_METHOD 0
122 POP_JUMP_IF_FALSE 104
178 124 LOAD_FAST 6 (tuple)
126 LOAD_CONST 6 (1)
128 BINARY_SUBSCR
130 LOAD_ATTR 1 (dtype)
132 ROT_TWO
134 POP_TOP
136 RETURN_VALUE
138 JUMP_ABSOLUTE 104
181 >> 140 LOAD_FAST 5 (last_tuple)
142 LOAD_CONST 6 (1)
144 BINARY_SUBSCR
146 LOAD_ATTR 1 (dtype)
148 RETURN_VALUE
150 LOAD_CONST 1 (None)
152 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST last_dtype [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST parameter []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR parameters [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [UserMethodVariable(<function Module.parameters at 0x7f3f1c0fb0d0>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE GET_ITER None [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 42 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST t [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST t [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST last_dtype [ListIteratorVariable(), TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST t [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_floating_point [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [ListIteratorVariable(), GetAttrVariable(TensorVariable(), is_floating_point)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 12 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST t [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE ROT_TWO None [ListIteratorVariable(), TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_TOP None [TorchVariable(torch.float32), ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object get_parameter_dtype at 0x7f3f1005d190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 153>
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object dtype at 0x7f3f100879d0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 727>
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dtype',) [GetAttrVariable(TensorVariable(), to), TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 1 [GetAttrVariable(TensorVariable(), to), TorchVariable(torch.float32), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST extended_attention_mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1.0 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBTRACT None [ConstantVariable(float), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -10000.0 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST extended_attention_mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object get_extended_attention_mask at 0x7f3f10087be0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 794>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST extended_attention_mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 590 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST encoder_extended_attention_mask [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR get_head_mask [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR num_layers [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object get_head_mask at 0x7f3f10087c90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 843>
861 0 LOAD_FAST 1 (head_mask)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 40
862 8 LOAD_FAST 0 (self)
10 LOAD_METHOD 0 (_convert_head_mask_to_5d)
12 LOAD_FAST 1 (head_mask)
14 LOAD_FAST 2 (num_hidden_layers)
16 CALL_METHOD 2
18 STORE_FAST 1 (head_mask)
863 20 LOAD_FAST 3 (is_attention_chunked)
22 LOAD_CONST 2 (True)
24 COMPARE_OP 8 (is)
26 POP_JUMP_IF_FALSE 50
864 28 LOAD_FAST 1 (head_mask)
30 LOAD_METHOD 1 (unsqueeze)
32 LOAD_CONST 3 (-1)
34 CALL_METHOD 1
36 STORE_FAST 1 (head_mask)
38 JUMP_FORWARD 10 (to 50)
866 >> 40 LOAD_CONST 1 (None)
42 BUILD_LIST 1
44 LOAD_FAST 2 (num_hidden_layers)
46 BINARY_MULTIPLY
48 STORE_FAST 1 (head_mask)
868 >> 50 LOAD_FAST 1 (head_mask)
52 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 40 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_LIST 1 [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_hidden_layers [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST head_mask [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object get_head_mask at 0x7f3f10087c90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 843>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST head_mask [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR get_head_mask [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR num_layers [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [UserMethodVariable(<function ModuleUtilsMixin.get_head_mask at 0x7f3ef770e310>, NNModuleVariable()), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object get_head_mask at 0x7f3f10087c90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 843>
861 0 LOAD_FAST 1 (head_mask)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 40
862 8 LOAD_FAST 0 (self)
10 LOAD_METHOD 0 (_convert_head_mask_to_5d)
12 LOAD_FAST 1 (head_mask)
14 LOAD_FAST 2 (num_hidden_layers)
16 CALL_METHOD 2
18 STORE_FAST 1 (head_mask)
863 20 LOAD_FAST 3 (is_attention_chunked)
22 LOAD_CONST 2 (True)
24 COMPARE_OP 8 (is)
26 POP_JUMP_IF_FALSE 50
864 28 LOAD_FAST 1 (head_mask)
30 LOAD_METHOD 1 (unsqueeze)
32 LOAD_CONST 3 (-1)
34 CALL_METHOD 1
36 STORE_FAST 1 (head_mask)
38 JUMP_FORWARD 10 (to 50)
866 >> 40 LOAD_CONST 1 (None)
42 BUILD_LIST 1
44 LOAD_FAST 2 (num_hidden_layers)
46 BINARY_MULTIPLY
48 STORE_FAST 1 (head_mask)
868 >> 50 LOAD_FAST 1 (head_mask)
52 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 40 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_LIST 1 [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_hidden_layers [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST head_mask [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object get_head_mask at 0x7f3f10087c90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/modeling_utils.py", line 843>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_head_mask [ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 636 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_states [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 650 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST all_hidden_states [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 664 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST all_attentions [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 686 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST all_cross_attentions [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST encoder_decoder_position_bias [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST inputs_embeds [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL enumerate []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL zip [BuiltinVariable(enumerate)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [BuiltinVariable(enumerate), BuiltinVariable(zip)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR block [BuiltinVariable(enumerate), BuiltinVariable(zip), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_values [BuiltinVariable(enumerate), BuiltinVariable(zip), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [BuiltinVariable(enumerate), BuiltinVariable(zip), NNModuleVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [BuiltinVariable(enumerate), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE GET_ITER None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR has_relative_attention_bias [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 324 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR compute_bias [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_length [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable()), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable()), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR device [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable()), ConstantVariable(int), ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('device',) [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable()), ConstantVariable(int), ConstantVariable(int), TorchVariable(cuda:0)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [UserMethodVariable(<function T5Attention.compute_bias at 0x7f3ef771c8b0>, NNModuleVariable()), ConstantVariable(int), ConstantVariable(int), TorchVariable(cuda:0), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object compute_bias at 0x7f3f100aeb30, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 419>
421 0 LOAD_FAST 3 (device)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 18
422 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (relative_attention_bias)
12 LOAD_ATTR 1 (weight)
14 LOAD_ATTR 2 (device)
16 STORE_FAST 3 (device)
423 >> 18 LOAD_GLOBAL 3 (torch)
20 LOAD_ATTR 4 (arange)
22 LOAD_FAST 1 (query_length)
24 LOAD_GLOBAL 3 (torch)
26 LOAD_ATTR 5 (long)
28 LOAD_FAST 3 (device)
30 LOAD_CONST 2 (('dtype', 'device'))
32 CALL_FUNCTION_KW 3
34 LOAD_CONST 1 (None)
36 LOAD_CONST 1 (None)
38 BUILD_SLICE 2
40 LOAD_CONST 1 (None)
42 BUILD_TUPLE 2
44 BINARY_SUBSCR
46 STORE_FAST 4 (context_position)
424 48 LOAD_GLOBAL 3 (torch)
50 LOAD_ATTR 4 (arange)
52 LOAD_FAST 2 (key_length)
54 LOAD_GLOBAL 3 (torch)
56 LOAD_ATTR 5 (long)
58 LOAD_FAST 3 (device)
60 LOAD_CONST 2 (('dtype', 'device'))
62 CALL_FUNCTION_KW 3
64 LOAD_CONST 1 (None)
66 LOAD_CONST 1 (None)
68 LOAD_CONST 1 (None)
70 BUILD_SLICE 2
72 BUILD_TUPLE 2
74 BINARY_SUBSCR
76 STORE_FAST 5 (memory_position)
425 78 LOAD_FAST 5 (memory_position)
80 LOAD_FAST 4 (context_position)
82 BINARY_SUBTRACT
84 STORE_FAST 6 (relative_position)
426 86 LOAD_FAST 0 (self)
88 LOAD_ATTR 6 (_relative_position_bucket)
427 90 LOAD_FAST 6 (relative_position)
428 92 LOAD_FAST 0 (self)
94 LOAD_ATTR 7 (is_decoder)
96 UNARY_NOT
429 98 LOAD_FAST 0 (self)
100 LOAD_ATTR 8 (relative_attention_num_buckets)
430 102 LOAD_FAST 0 (self)
104 LOAD_ATTR 9 (relative_attention_max_distance)
426 106 LOAD_CONST 3 (('bidirectional', 'num_buckets', 'max_distance'))
108 CALL_FUNCTION_KW 4
110 STORE_FAST 7 (relative_position_bucket)
432 112 LOAD_FAST 0 (self)
114 LOAD_METHOD 0 (relative_attention_bias)
116 LOAD_FAST 7 (relative_position_bucket)
118 CALL_METHOD 1
120 STORE_FAST 8 (values)
433 122 LOAD_FAST 8 (values)
124 LOAD_METHOD 10 (permute)
126 LOAD_CONST 4 (2)
128 LOAD_CONST 5 (0)
130 LOAD_CONST 6 (1)
132 BUILD_LIST 3
134 CALL_METHOD 1
136 LOAD_METHOD 11 (unsqueeze)
138 LOAD_CONST 5 (0)
140 CALL_METHOD 1
142 STORE_FAST 8 (values)
434 144 LOAD_FAST 8 (values)
146 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST device []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TorchVariable(cuda:0)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TorchVariable(cuda:0), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 18 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR arange [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_length [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR long [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST device [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dtype', 'device') [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype), TorchVariable(cuda:0)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype), TorchVariable(cuda:0), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TensorVariable(), SliceVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST context_position [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR arange [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_length [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR long [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST device [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dtype', 'device') [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype), TorchVariable(cuda:0)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<built-in method arange of type object at 0x7f40b0102e20>), ConstantVariable(int), ConstantVariable(dtype), TorchVariable(cuda:0), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TensorVariable(), ConstantVariable(NoneType), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST memory_position [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST memory_position []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST context_position [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBTRACT None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_position [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR _relative_position_bucket [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position [UserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [UserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNARY_NOT None [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR relative_attention_num_buckets [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR relative_attention_max_distance [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('bidirectional', 'num_buckets', 'max_distance') [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 4 [UserFunctionVariable(), TensorVariable(), ConstantVariable(bool), ConstantVariable(int), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object _relative_position_bucket at 0x7f3f100aea80, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 371>
393 0 LOAD_CONST 1 (0)
2 STORE_FAST 4 (relative_buckets)
394 4 LOAD_FAST 1 (bidirectional)
6 POP_JUMP_IF_FALSE 52
395 8 LOAD_FAST 2 (num_buckets)
10 LOAD_CONST 2 (2)
12 INPLACE_FLOOR_DIVIDE
14 STORE_FAST 2 (num_buckets)
396 16 LOAD_FAST 4 (relative_buckets)
18 LOAD_FAST 0 (relative_position)
20 LOAD_CONST 1 (0)
22 COMPARE_OP 4 (>)
24 LOAD_METHOD 0 (to)
26 LOAD_GLOBAL 1 (torch)
28 LOAD_ATTR 2 (long)
30 CALL_METHOD 1
32 LOAD_FAST 2 (num_buckets)
34 BINARY_MULTIPLY
36 INPLACE_ADD
38 STORE_FAST 4 (relative_buckets)
397 40 LOAD_GLOBAL 1 (torch)
42 LOAD_METHOD 3 (abs)
44 LOAD_FAST 0 (relative_position)
46 CALL_METHOD 1
48 STORE_FAST 0 (relative_position)
50 JUMP_FORWARD 20 (to 72)
399 >> 52 LOAD_GLOBAL 1 (torch)
54 LOAD_METHOD 4 (min)
56 LOAD_FAST 0 (relative_position)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_METHOD 5 (zeros_like)
62 LOAD_FAST 0 (relative_position)
64 CALL_METHOD 1
66 CALL_METHOD 2
68 UNARY_NEGATIVE
70 STORE_FAST 0 (relative_position)
403 >> 72 LOAD_FAST 2 (num_buckets)
74 LOAD_CONST 2 (2)
76 BINARY_FLOOR_DIVIDE
78 STORE_FAST 5 (max_exact)
404 80 LOAD_FAST 0 (relative_position)
82 LOAD_FAST 5 (max_exact)
84 COMPARE_OP 0 (<)
86 STORE_FAST 6 (is_small)
407 88 LOAD_FAST 5 (max_exact)
408 90 LOAD_GLOBAL 1 (torch)
92 LOAD_METHOD 6 (log)
94 LOAD_FAST 0 (relative_position)
96 LOAD_METHOD 7 (float)
98 CALL_METHOD 0
100 LOAD_FAST 5 (max_exact)
102 BINARY_TRUE_DIVIDE
104 CALL_METHOD 1
409 106 LOAD_GLOBAL 8 (math)
108 LOAD_METHOD 6 (log)
110 LOAD_FAST 3 (max_distance)
112 LOAD_FAST 5 (max_exact)
114 BINARY_TRUE_DIVIDE
116 CALL_METHOD 1
408 118 BINARY_TRUE_DIVIDE
410 120 LOAD_FAST 2 (num_buckets)
122 LOAD_FAST 5 (max_exact)
124 BINARY_SUBTRACT
408 126 BINARY_MULTIPLY
407 128 LOAD_METHOD 0 (to)
411 130 LOAD_GLOBAL 1 (torch)
132 LOAD_ATTR 2 (long)
407 134 CALL_METHOD 1
136 BINARY_ADD
138 STORE_FAST 7 (relative_position_if_large)
412 140 LOAD_GLOBAL 1 (torch)
142 LOAD_METHOD 4 (min)
413 144 LOAD_FAST 7 (relative_position_if_large)
146 LOAD_GLOBAL 1 (torch)
148 LOAD_METHOD 9 (full_like)
150 LOAD_FAST 7 (relative_position_if_large)
152 LOAD_FAST 2 (num_buckets)
154 LOAD_CONST 3 (1)
156 BINARY_SUBTRACT
158 CALL_METHOD 2
412 160 CALL_METHOD 2
162 STORE_FAST 7 (relative_position_if_large)
416 164 LOAD_FAST 4 (relative_buckets)
166 LOAD_GLOBAL 1 (torch)
168 LOAD_METHOD 10 (where)
170 LOAD_FAST 6 (is_small)
172 LOAD_FAST 0 (relative_position)
174 LOAD_FAST 7 (relative_position_if_large)
176 CALL_METHOD 3
178 INPLACE_ADD
180 STORE_FAST 4 (relative_buckets)
417 182 LOAD_FAST 4 (relative_buckets)
184 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_buckets [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST bidirectional []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 52 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_buckets []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_FLOOR_DIVIDE None [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST num_buckets [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_buckets []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP > [ConstantVariable(int), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [ConstantVariable(int), GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR long [ConstantVariable(int), GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [ConstantVariable(int), GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_buckets [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [ConstantVariable(int), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_buckets [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR abs [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position [TorchVariable(<built-in method abs of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TorchVariable(<built-in method abs of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_position [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 72 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_buckets []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_FLOOR_DIVIDE None [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST max_exact [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_exact [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP < [TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST is_small [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_exact []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR log [ConstantVariable(int), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_exact [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_TRUE_DIVIDE None [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [ConstantVariable(int), TorchVariable(<built-in method log of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL math [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR log [ConstantVariable(int), TensorVariable(), TorchVariable(<module 'math' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/lib-dynload/math.cpython-38-x86_64-linux-gnu.so'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_distance [ConstantVariable(int), TensorVariable(), TorchVariable(<built-in function log>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_exact [ConstantVariable(int), TensorVariable(), TorchVariable(<built-in function log>), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_TRUE_DIVIDE None [ConstantVariable(int), TensorVariable(), TorchVariable(<built-in function log>), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [ConstantVariable(int), TensorVariable(), TorchVariable(<built-in function log>), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_TRUE_DIVIDE None [ConstantVariable(int), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_buckets [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST max_exact [ConstantVariable(int), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBTRACT None [ConstantVariable(int), TensorVariable(), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [ConstantVariable(int), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [ConstantVariable(int), GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR long [ConstantVariable(int), GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [ConstantVariable(int), GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_position_if_large [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR min [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position_if_large [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR full_like [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position_if_large [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<built-in method full_like of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST num_buckets [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<built-in method full_like of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<built-in method full_like of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBTRACT None [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<built-in method full_like of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TorchVariable(<built-in method full_like of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method min of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_position_if_large [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_buckets []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR where [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST is_small [TensorVariable(), TorchVariable(<built-in method where of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position [TensorVariable(), TorchVariable(<built-in method where of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position_if_large [TensorVariable(), TorchVariable(<built-in method where of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [TensorVariable(), TorchVariable(<built-in method where of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_buckets [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_buckets []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object _relative_position_bucket at 0x7f3f100aea80, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 371>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST relative_position_bucket [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR relative_attention_bias [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST relative_position_bucket [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST values [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST values []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR permute [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), permute)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [GetAttrVariable(TensorVariable(), permute), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), permute), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_LIST 3 [GetAttrVariable(TensorVariable(), permute), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), permute), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR unsqueeze [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [GetAttrVariable(TensorVariable(), unsqueeze)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), unsqueeze), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST values [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST values []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object compute_bias at 0x7f3f100aeb30, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 419>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 392 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST mask [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST i [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_module [ListIteratorVariable(), ConstantVariable(NoneType), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST past_key_value [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_head_mask [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST i [ListIteratorVariable(), ListVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), ListVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_layer_head_mask [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 930 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 946 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR gradient_checkpointing [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1028 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_module [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_hidden_states [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_extended_attention_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_decoder_position_bias [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST cross_attn_layer_head_mask [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'encoder_hidden_states', 'encoder_attention_mask', 'encoder_decoder_position_bias', 'layer_head_mask', 'cross_attn_layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 11 [ListIteratorVariable(), NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
647 0 LOAD_FAST 9 (past_key_value)
2 LOAD_CONST 0 (None)
4 COMPARE_OP 9 (is not)
6 POP_JUMP_IF_FALSE 122
648 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (is_decoder)
12 POP_JUMP_IF_TRUE 24
649 14 LOAD_GLOBAL 1 (logger)
16 LOAD_METHOD 2 (warning)
18 LOAD_CONST 1 ('`past_key_values` is passed to the encoder. Please make sure this is intended.')
20 CALL_METHOD 1
22 POP_TOP
650 >> 24 LOAD_FAST 4 (encoder_hidden_states)
26 LOAD_CONST 0 (None)
28 COMPARE_OP 8 (is)
30 POP_JUMP_IF_FALSE 36
32 LOAD_CONST 2 (2)
34 JUMP_FORWARD 2 (to 38)
>> 36 LOAD_CONST 3 (4)
>> 38 STORE_FAST 13 (expected_num_past_key_values)
652 40 LOAD_GLOBAL 3 (len)
42 LOAD_FAST 9 (past_key_value)
44 CALL_FUNCTION 1
46 LOAD_FAST 13 (expected_num_past_key_values)
48 COMPARE_OP 3 (!=)
50 POP_JUMP_IF_FALSE 96
653 52 LOAD_GLOBAL 4 (ValueError)
654 54 LOAD_CONST 4 ('There should be ')
56 LOAD_FAST 13 (expected_num_past_key_values)
58 FORMAT_VALUE 0
60 LOAD_CONST 5 (' past states. ')
62 LOAD_FAST 13 (expected_num_past_key_values)
64 LOAD_CONST 3 (4)
66 COMPARE_OP 2 (==)
68 POP_JUMP_IF_FALSE 74
70 LOAD_CONST 6 ('2 (past / key) for cross attention. ')
72 JUMP_FORWARD 2 (to 76)
>> 74 LOAD_CONST 7 ('')
>> 76 FORMAT_VALUE 0
78 LOAD_CONST 8 ('Got ')
80 LOAD_GLOBAL 3 (len)
82 LOAD_FAST 9 (past_key_value)
84 CALL_FUNCTION 1
86 FORMAT_VALUE 0
88 LOAD_CONST 9 (' past key / value states')
90 BUILD_STRING 7
653 92 CALL_FUNCTION 1
94 RAISE_VARARGS 1
659 >> 96 LOAD_FAST 9 (past_key_value)
98 LOAD_CONST 0 (None)
100 LOAD_CONST 2 (2)
102 BUILD_SLICE 2
104 BINARY_SUBSCR
106 STORE_FAST 14 (self_attn_past_key_value)
660 108 LOAD_FAST 9 (past_key_value)
110 LOAD_CONST 2 (2)
112 LOAD_CONST 0 (None)
114 BUILD_SLICE 2
116 BINARY_SUBSCR
118 STORE_FAST 15 (cross_attn_past_key_value)
120 JUMP_FORWARD 8 (to 130)
662 >> 122 LOAD_CONST 10 ((None, None))
124 UNPACK_SEQUENCE 2
126 STORE_FAST 14 (self_attn_past_key_value)
128 STORE_FAST 15 (cross_attn_past_key_value)
664 >> 130 LOAD_FAST 0 (self)
132 LOAD_ATTR 5 (layer)
134 LOAD_CONST 11 (0)
136 BINARY_SUBSCR
665 138 LOAD_FAST 1 (hidden_states)
666 140 LOAD_FAST 2 (attention_mask)
667 142 LOAD_FAST 3 (position_bias)
668 144 LOAD_FAST 7 (layer_head_mask)
669 146 LOAD_FAST 14 (self_attn_past_key_value)
670 148 LOAD_FAST 10 (use_cache)
671 150 LOAD_FAST 11 (output_attentions)
664 152 LOAD_CONST 12 (('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
154 CALL_FUNCTION_KW 7
156 STORE_FAST 16 (self_attention_outputs)
673 158 LOAD_FAST 16 (self_attention_outputs)
160 LOAD_CONST 0 (None)
162 LOAD_CONST 2 (2)
164 BUILD_SLICE 2
166 BINARY_SUBSCR
168 UNPACK_SEQUENCE 2
170 STORE_FAST 1 (hidden_states)
172 STORE_FAST 17 (present_key_value_state)
674 174 LOAD_FAST 16 (self_attention_outputs)
176 LOAD_CONST 2 (2)
178 LOAD_CONST 0 (None)
180 BUILD_SLICE 2
182 BINARY_SUBSCR
184 STORE_FAST 18 (attention_outputs)
677 186 LOAD_FAST 1 (hidden_states)
188 LOAD_ATTR 6 (dtype)
190 LOAD_GLOBAL 7 (torch)
192 LOAD_ATTR 8 (float16)
194 COMPARE_OP 2 (==)
196 POP_JUMP_IF_FALSE 248
198 LOAD_GLOBAL 7 (torch)
200 LOAD_METHOD 9 (isinf)
202 LOAD_FAST 1 (hidden_states)
204 CALL_METHOD 1
206 LOAD_METHOD 10 (any)
208 CALL_METHOD 0
210 POP_JUMP_IF_FALSE 248
678 212 LOAD_GLOBAL 7 (torch)
214 LOAD_METHOD 11 (finfo)
216 LOAD_FAST 1 (hidden_states)
218 LOAD_ATTR 6 (dtype)
220 CALL_METHOD 1
222 LOAD_ATTR 12 (max)
224 LOAD_CONST 13 (1000)
226 BINARY_SUBTRACT
228 STORE_FAST 19 (clamp_value)
679 230 LOAD_GLOBAL 7 (torch)
232 LOAD_ATTR 13 (clamp)
234 LOAD_FAST 1 (hidden_states)
236 LOAD_FAST 19 (clamp_value)
238 UNARY_NEGATIVE
240 LOAD_FAST 19 (clamp_value)
242 LOAD_CONST 14 (('min', 'max'))
244 CALL_FUNCTION_KW 3
246 STORE_FAST 1 (hidden_states)
681 >> 248 LOAD_FAST 0 (self)
250 LOAD_ATTR 0 (is_decoder)
252 EXTENDED_ARG 1
254 JUMP_IF_FALSE_OR_POP 262
256 LOAD_FAST 4 (encoder_hidden_states)
258 LOAD_CONST 0 (None)
260 COMPARE_OP 9 (is not)
>> 262 STORE_FAST 20 (do_cross_attention)
682 264 LOAD_FAST 20 (do_cross_attention)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_FALSE 444
685 270 LOAD_FAST 17 (present_key_value_state)
272 LOAD_CONST 0 (None)
274 COMPARE_OP 9 (is not)
276 EXTENDED_ARG 1
278 POP_JUMP_IF_FALSE 296
686 280 LOAD_FAST 17 (present_key_value_state)
282 LOAD_CONST 11 (0)
284 BINARY_SUBSCR
286 LOAD_ATTR 14 (shape)
288 LOAD_CONST 2 (2)
290 BINARY_SUBSCR
292 STORE_FAST 21 (query_length)
294 JUMP_FORWARD 4 (to 300)
688 >> 296 LOAD_CONST 0 (None)
298 STORE_FAST 21 (query_length)
690 >> 300 LOAD_FAST 0 (self)
302 LOAD_ATTR 5 (layer)
304 LOAD_CONST 15 (1)
306 BINARY_SUBSCR
691 308 LOAD_FAST 1 (hidden_states)
692 310 LOAD_FAST 4 (encoder_hidden_states)
693 312 LOAD_FAST 5 (encoder_attention_mask)
694 314 LOAD_FAST 6 (encoder_decoder_position_bias)
695 316 LOAD_FAST 8 (cross_attn_layer_head_mask)
696 318 LOAD_FAST 15 (cross_attn_past_key_value)
697 320 LOAD_FAST 21 (query_length)
698 322 LOAD_FAST 10 (use_cache)
699 324 LOAD_FAST 11 (output_attentions)
690 326 LOAD_CONST 16 (('key_value_states', 'attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'query_length', 'use_cache', 'output_attentions'))
328 CALL_FUNCTION_KW 9
330 STORE_FAST 22 (cross_attention_outputs)
701 332 LOAD_FAST 22 (cross_attention_outputs)
334 LOAD_CONST 11 (0)
336 BINARY_SUBSCR
338 STORE_FAST 1 (hidden_states)
704 340 LOAD_FAST 1 (hidden_states)
342 LOAD_ATTR 6 (dtype)
344 LOAD_GLOBAL 7 (torch)
346 LOAD_ATTR 8 (float16)
348 COMPARE_OP 2 (==)
350 EXTENDED_ARG 1
352 POP_JUMP_IF_FALSE 406
354 LOAD_GLOBAL 7 (torch)
356 LOAD_METHOD 9 (isinf)
358 LOAD_FAST 1 (hidden_states)
360 CALL_METHOD 1
362 LOAD_METHOD 10 (any)
364 CALL_METHOD 0
366 EXTENDED_ARG 1
368 POP_JUMP_IF_FALSE 406
705 370 LOAD_GLOBAL 7 (torch)
372 LOAD_METHOD 11 (finfo)
374 LOAD_FAST 1 (hidden_states)
376 LOAD_ATTR 6 (dtype)
378 CALL_METHOD 1
380 LOAD_ATTR 12 (max)
382 LOAD_CONST 13 (1000)
384 BINARY_SUBTRACT
386 STORE_FAST 19 (clamp_value)
706 388 LOAD_GLOBAL 7 (torch)
390 LOAD_ATTR 13 (clamp)
392 LOAD_FAST 1 (hidden_states)
394 LOAD_FAST 19 (clamp_value)
396 UNARY_NEGATIVE
398 LOAD_FAST 19 (clamp_value)
400 LOAD_CONST 14 (('min', 'max'))
402 CALL_FUNCTION_KW 3
404 STORE_FAST 1 (hidden_states)
709 >> 406 LOAD_FAST 17 (present_key_value_state)
408 LOAD_CONST 0 (None)
410 COMPARE_OP 9 (is not)
412 EXTENDED_ARG 1
414 POP_JUMP_IF_FALSE 428
710 416 LOAD_FAST 17 (present_key_value_state)
418 LOAD_FAST 22 (cross_attention_outputs)
420 LOAD_CONST 15 (1)
422 BINARY_SUBSCR
424 BINARY_ADD
426 STORE_FAST 17 (present_key_value_state)
713 >> 428 LOAD_FAST 18 (attention_outputs)
430 LOAD_FAST 22 (cross_attention_outputs)
432 LOAD_CONST 2 (2)
434 LOAD_CONST 0 (None)
436 BUILD_SLICE 2
438 BINARY_SUBSCR
440 BINARY_ADD
442 STORE_FAST 18 (attention_outputs)
716 >> 444 LOAD_FAST 0 (self)
446 LOAD_ATTR 5 (layer)
448 LOAD_CONST 17 (-1)
450 BINARY_SUBSCR
452 LOAD_FAST 1 (hidden_states)
454 CALL_FUNCTION 1
456 STORE_FAST 1 (hidden_states)
719 458 LOAD_FAST 1 (hidden_states)
460 LOAD_ATTR 6 (dtype)
462 LOAD_GLOBAL 7 (torch)
464 LOAD_ATTR 8 (float16)
466 COMPARE_OP 2 (==)
468 EXTENDED_ARG 2
470 POP_JUMP_IF_FALSE 524
472 LOAD_GLOBAL 7 (torch)
474 LOAD_METHOD 9 (isinf)
476 LOAD_FAST 1 (hidden_states)
478 CALL_METHOD 1
480 LOAD_METHOD 10 (any)
482 CALL_METHOD 0
484 EXTENDED_ARG 2
486 POP_JUMP_IF_FALSE 524
720 488 LOAD_GLOBAL 7 (torch)
490 LOAD_METHOD 11 (finfo)
492 LOAD_FAST 1 (hidden_states)
494 LOAD_ATTR 6 (dtype)
496 CALL_METHOD 1
498 LOAD_ATTR 12 (max)
500 LOAD_CONST 13 (1000)
502 BINARY_SUBTRACT
504 STORE_FAST 19 (clamp_value)
721 506 LOAD_GLOBAL 7 (torch)
508 LOAD_ATTR 13 (clamp)
510 LOAD_FAST 1 (hidden_states)
512 LOAD_FAST 19 (clamp_value)
514 UNARY_NEGATIVE
516 LOAD_FAST 19 (clamp_value)
518 LOAD_CONST 14 (('min', 'max'))
520 CALL_FUNCTION_KW 3
522 STORE_FAST 1 (hidden_states)
723 >> 524 LOAD_FAST 1 (hidden_states)
526 BUILD_TUPLE 1
528 STORE_FAST 23 (outputs)
725 530 LOAD_FAST 10 (use_cache)
532 EXTENDED_ARG 2
534 POP_JUMP_IF_FALSE 552
726 536 LOAD_FAST 23 (outputs)
538 LOAD_FAST 17 (present_key_value_state)
540 BUILD_TUPLE 1
542 BINARY_ADD
544 LOAD_FAST 18 (attention_outputs)
546 BINARY_ADD
548 STORE_FAST 23 (outputs)
550 JUMP_FORWARD 8 (to 560)
728 >> 552 LOAD_FAST 23 (outputs)
554 LOAD_FAST 18 (attention_outputs)
556 BINARY_ADD
558 STORE_FAST 23 (outputs)
730 >> 560 LOAD_FAST 23 (outputs)
562 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 122 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None, None) []
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attn_past_key_value [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST cross_attn_past_key_value [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attn_past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('attention_mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
569 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 8 (normed_hidden_states)
570 10 LOAD_FAST 0 (self)
12 LOAD_ATTR 1 (SelfAttention)
571 14 LOAD_FAST 8 (normed_hidden_states)
572 16 LOAD_FAST 2 (attention_mask)
573 18 LOAD_FAST 3 (position_bias)
574 20 LOAD_FAST 4 (layer_head_mask)
575 22 LOAD_FAST 5 (past_key_value)
576 24 LOAD_FAST 6 (use_cache)
577 26 LOAD_FAST 7 (output_attentions)
570 28 LOAD_CONST 1 (('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions'))
30 CALL_FUNCTION_KW 7
32 STORE_FAST 9 (attention_output)
579 34 LOAD_FAST 1 (hidden_states)
36 LOAD_FAST 0 (self)
38 LOAD_METHOD 2 (dropout)
40 LOAD_FAST 9 (attention_output)
42 LOAD_CONST 2 (0)
44 BINARY_SUBSCR
46 CALL_METHOD 1
48 BINARY_ADD
50 STORE_FAST 1 (hidden_states)
580 52 LOAD_FAST 1 (hidden_states)
54 BUILD_TUPLE 1
56 LOAD_FAST 9 (attention_output)
58 LOAD_CONST 3 (1)
60 LOAD_CONST 0 (None)
62 BUILD_SLICE 2
64 BINARY_SUBSCR
66 BINARY_ADD
68 STORE_FAST 10 (outputs)
581 70 LOAD_FAST 10 (outputs)
72 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST normed_hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR SelfAttention [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST normed_hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_mask [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [NNModuleVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('mask', 'position_bias', 'layer_head_mask', 'past_key_value', 'use_cache', 'output_attentions') [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 7 [NNModuleVariable(), TensorVariable(), TensorVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(bool), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
454 0 LOAD_FAST 1 (hidden_states)
2 LOAD_ATTR 0 (shape)
4 LOAD_CONST 1 (None)
6 LOAD_CONST 2 (2)
8 BUILD_SLICE 2
10 BINARY_SUBSCR
12 UNPACK_SEQUENCE 2
14 STORE_DEREF 0 (batch_size)
16 STORE_FAST 10 (seq_length)
456 18 LOAD_FAST 10 (seq_length)
20 STORE_FAST 11 (real_seq_length)
458 22 LOAD_FAST 5 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 9 (is not)
28 POP_JUMP_IF_FALSE 92
460 30 LOAD_GLOBAL 1 (len)
32 LOAD_FAST 5 (past_key_value)
34 CALL_FUNCTION 1
36 LOAD_CONST 2 (2)
38 COMPARE_OP 2 (==)
459 40 POP_JUMP_IF_TRUE 62
42 LOAD_GLOBAL 2 (AssertionError)
461 44 LOAD_CONST 3 ('past_key_value should have 2 past states: keys and values. Got ')
46 LOAD_GLOBAL 1 (len)
48 LOAD_FAST 5 (past_key_value)
50 CALL_FUNCTION 1
52 FORMAT_VALUE 0
54 LOAD_CONST 4 (' past states')
56 BUILD_STRING 3
459 58 CALL_FUNCTION 1
60 RAISE_VARARGS 1
462 >> 62 LOAD_FAST 11 (real_seq_length)
64 LOAD_FAST 7 (query_length)
66 LOAD_CONST 1 (None)
68 COMPARE_OP 8 (is)
70 POP_JUMP_IF_FALSE 86
72 LOAD_FAST 5 (past_key_value)
74 LOAD_CONST 5 (0)
76 BINARY_SUBSCR
78 LOAD_ATTR 0 (shape)
80 LOAD_CONST 2 (2)
82 BINARY_SUBSCR
84 JUMP_FORWARD 2 (to 88)
>> 86 LOAD_FAST 7 (query_length)
>> 88 INPLACE_ADD
90 STORE_FAST 11 (real_seq_length)
464 >> 92 LOAD_FAST 3 (key_value_states)
94 LOAD_CONST 1 (None)
96 COMPARE_OP 8 (is)
98 POP_JUMP_IF_FALSE 104
100 LOAD_FAST 11 (real_seq_length)
102 JUMP_FORWARD 8 (to 112)
>> 104 LOAD_FAST 3 (key_value_states)
106 LOAD_ATTR 0 (shape)
108 LOAD_CONST 6 (1)
110 BINARY_SUBSCR
>> 112 STORE_FAST 12 (key_length)
466 114 LOAD_CLOSURE 0 (batch_size)
116 LOAD_CLOSURE 1 (self)
118 BUILD_TUPLE 2
120 LOAD_CONST 7 (<code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>)
122 LOAD_CONST 8 ('T5Attention.forward.<locals>.shape')
124 MAKE_FUNCTION 8 (closure)
126 STORE_DEREF 2 (shape)
470 128 LOAD_CLOSURE 0 (batch_size)
130 LOAD_CLOSURE 1 (self)
132 BUILD_TUPLE 2
134 LOAD_CONST 9 (<code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>)
136 LOAD_CONST 10 ('T5Attention.forward.<locals>.unshape')
138 MAKE_FUNCTION 8 (closure)
140 STORE_FAST 13 (unshape)
474 142 LOAD_CLOSURE 2 (shape)
144 BUILD_TUPLE 1
146 LOAD_CONST 11 (<code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>)
148 LOAD_CONST 12 ('T5Attention.forward.<locals>.project')
150 MAKE_FUNCTION 8 (closure)
152 STORE_FAST 14 (project)
496 154 LOAD_DEREF 2 (shape)
156 LOAD_DEREF 1 (self)
158 LOAD_METHOD 3 (q)
160 LOAD_FAST 1 (hidden_states)
162 CALL_METHOD 1
164 CALL_FUNCTION 1
166 STORE_FAST 15 (query_states)
499 168 LOAD_FAST 14 (project)
500 170 LOAD_FAST 1 (hidden_states)
172 LOAD_DEREF 1 (self)
174 LOAD_ATTR 4 (k)
176 LOAD_FAST 3 (key_value_states)
178 LOAD_FAST 5 (past_key_value)
180 LOAD_CONST 1 (None)
182 COMPARE_OP 9 (is not)
184 POP_JUMP_IF_FALSE 194
186 LOAD_FAST 5 (past_key_value)
188 LOAD_CONST 5 (0)
190 BINARY_SUBSCR
192 JUMP_FORWARD 2 (to 196)
>> 194 LOAD_CONST 1 (None)
499 >> 196 CALL_FUNCTION 4
198 STORE_FAST 16 (key_states)
502 200 LOAD_FAST 14 (project)
503 202 LOAD_FAST 1 (hidden_states)
204 LOAD_DEREF 1 (self)
206 LOAD_ATTR 5 (v)
208 LOAD_FAST 3 (key_value_states)
210 LOAD_FAST 5 (past_key_value)
212 LOAD_CONST 1 (None)
214 COMPARE_OP 9 (is not)
216 POP_JUMP_IF_FALSE 226
218 LOAD_FAST 5 (past_key_value)
220 LOAD_CONST 6 (1)
222 BINARY_SUBSCR
224 JUMP_FORWARD 2 (to 228)
>> 226 LOAD_CONST 1 (None)
502 >> 228 CALL_FUNCTION 4
230 STORE_FAST 17 (value_states)
507 232 LOAD_GLOBAL 6 (torch)
234 LOAD_METHOD 7 (matmul)
508 236 LOAD_FAST 15 (query_states)
238 LOAD_FAST 16 (key_states)
240 LOAD_METHOD 8 (transpose)
242 LOAD_CONST 13 (3)
244 LOAD_CONST 2 (2)
246 CALL_METHOD 2
507 248 CALL_METHOD 2
250 STORE_FAST 18 (scores)
511 252 LOAD_FAST 4 (position_bias)
254 LOAD_CONST 1 (None)
256 COMPARE_OP 8 (is)
258 EXTENDED_ARG 1
260 POP_JUMP_IF_FALSE 410
512 262 LOAD_DEREF 1 (self)
264 LOAD_ATTR 9 (has_relative_attention_bias)
266 EXTENDED_ARG 1
268 POP_JUMP_IF_TRUE 324
513 270 LOAD_GLOBAL 6 (torch)
272 LOAD_ATTR 10 (zeros)
514 274 LOAD_CONST 6 (1)
276 LOAD_DEREF 1 (self)
278 LOAD_ATTR 11 (n_heads)
280 LOAD_FAST 11 (real_seq_length)
282 LOAD_FAST 12 (key_length)
284 BUILD_TUPLE 4
286 LOAD_FAST 18 (scores)
288 LOAD_ATTR 12 (device)
290 LOAD_FAST 18 (scores)
292 LOAD_ATTR 13 (dtype)
513 294 LOAD_CONST 14 (('device', 'dtype'))
296 CALL_FUNCTION_KW 3
298 STORE_FAST 4 (position_bias)
516 300 LOAD_DEREF 1 (self)
302 LOAD_ATTR 14 (gradient_checkpointing)
304 EXTENDED_ARG 1
306 POP_JUMP_IF_FALSE 342
308 LOAD_DEREF 1 (self)
310 LOAD_ATTR 15 (training)
312 EXTENDED_ARG 1
314 POP_JUMP_IF_FALSE 342
517 316 LOAD_CONST 15 (True)
318 LOAD_FAST 4 (position_bias)
320 STORE_ATTR 16 (requires_grad)
322 JUMP_FORWARD 18 (to 342)
519 >> 324 LOAD_DEREF 1 (self)
326 LOAD_ATTR 17 (compute_bias)
328 LOAD_FAST 11 (real_seq_length)
330 LOAD_FAST 12 (key_length)
332 LOAD_FAST 18 (scores)
334 LOAD_ATTR 12 (device)
336 LOAD_CONST 16 (('device',))
338 CALL_FUNCTION_KW 3
340 STORE_FAST 4 (position_bias)
523 >> 342 LOAD_FAST 5 (past_key_value)
344 LOAD_CONST 1 (None)
346 COMPARE_OP 9 (is not)
348 EXTENDED_ARG 1
350 POP_JUMP_IF_FALSE 392
524 352 LOAD_FAST 4 (position_bias)
354 LOAD_CONST 1 (None)
356 LOAD_CONST 1 (None)
358 BUILD_SLICE 2
360 LOAD_CONST 1 (None)
362 LOAD_CONST 1 (None)
364 BUILD_SLICE 2
366 LOAD_FAST 1 (hidden_states)
368 LOAD_METHOD 18 (size)
370 LOAD_CONST 6 (1)
372 CALL_METHOD 1
374 UNARY_NEGATIVE
376 LOAD_CONST 1 (None)
378 BUILD_SLICE 2
380 LOAD_CONST 1 (None)
382 LOAD_CONST 1 (None)
384 BUILD_SLICE 2
386 BUILD_TUPLE 4
388 BINARY_SUBSCR
390 STORE_FAST 4 (position_bias)
526 >> 392 LOAD_FAST 2 (mask)
394 LOAD_CONST 1 (None)
396 COMPARE_OP 9 (is not)
398 EXTENDED_ARG 1
400 POP_JUMP_IF_FALSE 410
527 402 LOAD_FAST 4 (position_bias)
404 LOAD_FAST 2 (mask)
406 BINARY_ADD
408 STORE_FAST 4 (position_bias)
529 >> 410 LOAD_FAST 18 (scores)
412 LOAD_FAST 4 (position_bias)
414 INPLACE_ADD
416 STORE_FAST 18 (scores)
530 418 LOAD_GLOBAL 19 (nn)
420 LOAD_ATTR 20 (functional)
422 LOAD_ATTR 21 (softmax)
424 LOAD_FAST 18 (scores)
426 LOAD_METHOD 22 (float)
428 CALL_METHOD 0
430 LOAD_CONST 17 (-1)
432 LOAD_CONST 18 (('dim',))
434 CALL_FUNCTION_KW 2
436 LOAD_METHOD 23 (type_as)
531 438 LOAD_FAST 18 (scores)
530 440 CALL_METHOD 1
442 STORE_FAST 19 (attn_weights)
533 444 LOAD_GLOBAL 19 (nn)
446 LOAD_ATTR 20 (functional)
448 LOAD_ATTR 24 (dropout)
534 450 LOAD_FAST 19 (attn_weights)
452 LOAD_DEREF 1 (self)
454 LOAD_ATTR 24 (dropout)
456 LOAD_DEREF 1 (self)
458 LOAD_ATTR 15 (training)
533 460 LOAD_CONST 19 (('p', 'training'))
462 CALL_FUNCTION_KW 3
464 STORE_FAST 19 (attn_weights)
538 466 LOAD_FAST 6 (layer_head_mask)
468 LOAD_CONST 1 (None)
470 COMPARE_OP 9 (is not)
472 EXTENDED_ARG 1
474 POP_JUMP_IF_FALSE 484
539 476 LOAD_FAST 19 (attn_weights)
478 LOAD_FAST 6 (layer_head_mask)
480 BINARY_MULTIPLY
482 STORE_FAST 19 (attn_weights)
541 >> 484 LOAD_FAST 13 (unshape)
486 LOAD_GLOBAL 6 (torch)
488 LOAD_METHOD 7 (matmul)
490 LOAD_FAST 19 (attn_weights)
492 LOAD_FAST 17 (value_states)
494 CALL_METHOD 2
496 CALL_FUNCTION 1
498 STORE_FAST 20 (attn_output)
542 500 LOAD_DEREF 1 (self)
502 LOAD_METHOD 25 (o)
504 LOAD_FAST 20 (attn_output)
506 CALL_METHOD 1
508 STORE_FAST 20 (attn_output)
544 510 LOAD_DEREF 1 (self)
512 LOAD_ATTR 26 (is_decoder)
514 EXTENDED_ARG 2
516 POP_JUMP_IF_FALSE 532
518 LOAD_FAST 8 (use_cache)
520 EXTENDED_ARG 2
522 POP_JUMP_IF_FALSE 532
524 LOAD_FAST 16 (key_states)
526 LOAD_FAST 17 (value_states)
528 BUILD_TUPLE 2
530 JUMP_FORWARD 2 (to 534)
>> 532 LOAD_CONST 1 (None)
>> 534 STORE_FAST 21 (present_key_value_state)
545 536 LOAD_FAST 20 (attn_output)
538 BUILD_TUPLE 1
540 LOAD_FAST 21 (present_key_value_state)
542 BUILD_TUPLE 1
544 BINARY_ADD
546 LOAD_FAST 4 (position_bias)
548 BUILD_TUPLE 1
550 BINARY_ADD
552 STORE_FAST 22 (outputs)
547 554 LOAD_FAST 9 (output_attentions)
556 EXTENDED_ARG 2
558 POP_JUMP_IF_FALSE 570
548 560 LOAD_FAST 22 (outputs)
562 LOAD_FAST 19 (attn_weights)
564 BUILD_TUPLE 1
566 BINARY_ADD
568 STORE_FAST 22 (outputs)
549 >> 570 LOAD_FAST 22 (outputs)
572 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ShapeVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ShapeVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ShapeVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF batch_size [ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST real_seq_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 92 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 104 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST real_seq_length []
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 112 [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_length [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.shape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_DEREF shape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE batch_size []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE self [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [NewCellVariable(), NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.unshape [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST unshape [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CLOSURE shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [NewCellVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474> [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST T5Attention.forward.<locals>.project [TupleVariable(), ConstantVariable(code)]
torchdynamo.symbolic_convert: [DEBUG] TRACE MAKE_FUNCTION 8 [TupleVariable(), ConstantVariable(code), ConstantVariable(str)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST project [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR q [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST query_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR k [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 194 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST key_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST project []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR v [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 226 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [NestedUserFunctionVariable(), TensorVariable(), NNModuleVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
476 0 LOAD_FAST 2 (key_value_states)
2 LOAD_CONST 1 (None)
4 COMPARE_OP 8 (is)
6 POP_JUMP_IF_FALSE 22
479 8 LOAD_DEREF 0 (shape)
10 LOAD_FAST 1 (proj_layer)
12 LOAD_FAST 0 (hidden_states)
14 CALL_FUNCTION 1
16 CALL_FUNCTION 1
18 STORE_FAST 0 (hidden_states)
20 JUMP_FORWARD 20 (to 42)
480 >> 22 LOAD_FAST 3 (past_key_value)
24 LOAD_CONST 1 (None)
26 COMPARE_OP 8 (is)
28 POP_JUMP_IF_FALSE 42
483 30 LOAD_DEREF 0 (shape)
32 LOAD_FAST 1 (proj_layer)
34 LOAD_FAST 2 (key_value_states)
36 CALL_FUNCTION 1
38 CALL_FUNCTION 1
40 STORE_FAST 0 (hidden_states)
485 >> 42 LOAD_FAST 3 (past_key_value)
44 LOAD_CONST 1 (None)
46 COMPARE_OP 9 (is not)
48 POP_JUMP_IF_FALSE 82
486 50 LOAD_FAST 2 (key_value_states)
52 LOAD_CONST 1 (None)
54 COMPARE_OP 8 (is)
56 POP_JUMP_IF_FALSE 78
489 58 LOAD_GLOBAL 0 (torch)
60 LOAD_ATTR 1 (cat)
62 LOAD_FAST 3 (past_key_value)
64 LOAD_FAST 0 (hidden_states)
66 BUILD_LIST 2
68 LOAD_CONST 2 (2)
70 LOAD_CONST 3 (('dim',))
72 CALL_FUNCTION_KW 2
74 STORE_FAST 0 (hidden_states)
76 JUMP_FORWARD 4 (to 82)
492 >> 78 LOAD_FAST 3 (past_key_value)
80 STORE_FAST 0 (hidden_states)
493 >> 82 LOAD_FAST 0 (hidden_states)
84 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_value_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 22 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF shape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST proj_layer [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NestedUserFunctionVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
468 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (view)
4 LOAD_DEREF 0 (batch_size)
6 LOAD_CONST 1 (-1)
8 LOAD_DEREF 1 (self)
10 LOAD_ATTR 1 (n_heads)
12 LOAD_DEREF 1 (self)
14 LOAD_ATTR 2 (key_value_proj_dim)
16 CALL_METHOD 4
18 LOAD_METHOD 3 (transpose)
20 LOAD_CONST 2 (1)
22 LOAD_CONST 3 (2)
24 CALL_METHOD 2
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR n_heads [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR key_value_proj_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 4 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object shape at 0x7f3f100aebe0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 466>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 42 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST past_key_value []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object project at 0x7f3f100aed40, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 474>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST value_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST query_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST key_states [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 3 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 410 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE INPLACE_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST scores [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR softmax [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [TorchVariable(<function softmax at 0x7f3f1bf53e50>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), GetAttrVariable(TensorVariable(), float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('dim',) [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [TorchVariable(<function softmax at 0x7f3f1bf53e50>), TensorVariable(), ConstantVariable(int), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR type_as [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST scores [GetAttrVariable(TensorVariable(), type_as)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), type_as), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL nn []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR functional [TorchVariable(<module 'torch.nn' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<module 'torch.nn.functional' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/nn/functional.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [TorchVariable(<function dropout at 0x7f3f1bf53310>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR training [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('p', 'training') [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 3 [TorchVariable(<function dropout at 0x7f3f1bf53310>), TensorVariable(), ConstantVariable(float), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_weights [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_head_mask []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 484 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST unshape []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [NestedUserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR matmul [NestedUserFunctionVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_weights [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST value_states [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [NestedUserFunctionVariable(), TorchVariable(<built-in method matmul of type object at 0x7f40b0102e20>), TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NestedUserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
472 0 LOAD_FAST 0 (states)
2 LOAD_METHOD 0 (transpose)
4 LOAD_CONST 1 (1)
6 LOAD_CONST 2 (2)
8 CALL_METHOD 2
10 LOAD_METHOD 1 (contiguous)
12 CALL_METHOD 0
14 LOAD_METHOD 2 (view)
16 LOAD_DEREF 0 (batch_size)
18 LOAD_CONST 3 (-1)
20 LOAD_DEREF 1 (self)
22 LOAD_ATTR 3 (inner_dim)
24 CALL_METHOD 3
26 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR transpose [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [GetAttrVariable(TensorVariable(), transpose)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), transpose), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR contiguous [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), contiguous)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR view [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF batch_size [GetAttrVariable(TensorVariable(), view)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR inner_dim [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 3 [GetAttrVariable(TensorVariable(), view), ConstantVariable(int), ConstantVariable(int), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object unshape at 0x7f3f100aec90, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 470>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR o [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attn_output [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 532 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None []
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attn_output []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_state [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST position_bias [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TupleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_attentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 570 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100aedf0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 436>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TensorVariable(), NNModuleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), NNModuleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_output [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 559>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST self_attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self_attention_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 248 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_IF_FALSE_OR_POP 262 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST do_cross_attention [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST do_cross_attention []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 444 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [NNModuleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
325 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (layer_norm)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 2 (forwarded_states)
326 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (DenseReluDense)
14 LOAD_FAST 2 (forwarded_states)
16 CALL_METHOD 1
18 STORE_FAST 2 (forwarded_states)
327 20 LOAD_FAST 1 (hidden_states)
22 LOAD_FAST 0 (self)
24 LOAD_METHOD 2 (dropout)
26 LOAD_FAST 2 (forwarded_states)
28 CALL_METHOD 1
30 BINARY_ADD
32 STORE_FAST 1 (hidden_states)
328 34 LOAD_FAST 1 (hidden_states)
36 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR DenseReluDense [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
288 0 LOAD_FAST 0 (self)
2 LOAD_METHOD 0 (wi)
4 LOAD_FAST 1 (hidden_states)
6 CALL_METHOD 1
8 STORE_FAST 1 (hidden_states)
289 10 LOAD_FAST 0 (self)
12 LOAD_METHOD 1 (act)
14 LOAD_FAST 1 (hidden_states)
16 CALL_METHOD 1
18 STORE_FAST 1 (hidden_states)
290 20 LOAD_FAST 0 (self)
22 LOAD_METHOD 2 (dropout)
24 LOAD_FAST 1 (hidden_states)
26 CALL_METHOD 1
28 STORE_FAST 1 (hidden_states)
291 30 LOAD_FAST 0 (self)
32 LOAD_METHOD 3 (wo)
34 LOAD_FAST 1 (hidden_states)
36 CALL_METHOD 1
38 STORE_FAST 1 (hidden_states)
292 40 LOAD_FAST 1 (hidden_states)
42 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wi [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR act [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR wo [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae3a0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 287>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST forwarded_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST forwarded_states [TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae7c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 324>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 524 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 1 [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST use_cache []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 552 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST attention_outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST outputs [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100b4450, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 631>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST False [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ListIteratorVariable(), ConstantVariable(bool), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1096 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (None,) [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), TupleVariable(), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [ListIteratorVariable(), TupleVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST layer_outputs [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [ListIteratorVariable(), TupleVariable(), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE UNPACK_SEQUENCE 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [ListIteratorVariable(), ConstantVariable(NoneType), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST present_key_value_state [ListIteratorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST layer_outputs [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [ListIteratorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [ListIteratorVariable(), TupleVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST position_bias [ListIteratorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR is_decoder [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1156 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF use_cache [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1172 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_DEREF output_attentions [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1214 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [ListIteratorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 724 [ListIteratorVariable(), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE FOR_ITER 1304 [ListIteratorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR final_layer_norm [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
255 0 LOAD_FAST 1 (hidden_states)
2 LOAD_METHOD 0 (to)
4 LOAD_GLOBAL 1 (torch)
6 LOAD_ATTR 2 (float32)
8 CALL_METHOD 1
10 LOAD_METHOD 3 (pow)
12 LOAD_CONST 1 (2)
14 CALL_METHOD 1
16 LOAD_ATTR 4 (mean)
18 LOAD_CONST 2 (-1)
20 LOAD_CONST 3 (True)
22 LOAD_CONST 4 (('keepdim',))
24 CALL_FUNCTION_KW 2
26 STORE_FAST 2 (variance)
256 28 LOAD_FAST 1 (hidden_states)
30 LOAD_GLOBAL 1 (torch)
32 LOAD_METHOD 5 (rsqrt)
34 LOAD_FAST 2 (variance)
36 LOAD_FAST 0 (self)
38 LOAD_ATTR 6 (variance_epsilon)
40 BINARY_ADD
42 CALL_METHOD 1
44 BINARY_MULTIPLY
46 STORE_FAST 1 (hidden_states)
259 48 LOAD_FAST 0 (self)
50 LOAD_ATTR 7 (weight)
52 LOAD_ATTR 8 (dtype)
54 LOAD_GLOBAL 1 (torch)
56 LOAD_ATTR 9 (float16)
58 LOAD_GLOBAL 1 (torch)
60 LOAD_ATTR 10 (bfloat16)
62 BUILD_TUPLE 2
64 COMPARE_OP 6 (in)
66 POP_JUMP_IF_FALSE 82
260 68 LOAD_FAST 1 (hidden_states)
70 LOAD_METHOD 0 (to)
72 LOAD_FAST 0 (self)
74 LOAD_ATTR 7 (weight)
76 LOAD_ATTR 8 (dtype)
78 CALL_METHOD 1
80 STORE_FAST 1 (hidden_states)
262 >> 82 LOAD_FAST 0 (self)
84 LOAD_ATTR 7 (weight)
86 LOAD_FAST 1 (hidden_states)
88 BINARY_MULTIPLY
90 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR to [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [GetAttrVariable(TensorVariable(), to)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float32 [GetAttrVariable(TensorVariable(), to), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), to), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pow [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 2 [GetAttrVariable(TensorVariable(), pow)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), pow), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR mean [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [GetAttrVariable(TensorVariable(), mean)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST True [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('keepdim',) [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 2 [GetAttrVariable(TensorVariable(), mean), ConstantVariable(int), ConstantVariable(bool), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST variance [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR rsqrt [TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST variance [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR variance_epsilon [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_ADD None [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(float)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TensorVariable(), TorchVariable(<built-in method rsqrt of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dtype [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR float16 [TorchVariable(torch.float32), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch [TorchVariable(torch.float32), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR bfloat16 [TorchVariable(torch.float32), ConstantVariable(dtype), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TorchVariable(torch.float32), ConstantVariable(dtype), ConstantVariable(dtype)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP in [TorchVariable(torch.float32), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 82 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR weight [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_MULTIPLY None [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ae190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 248>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR dropout [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST output_hidden_states []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 1340 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST return_dict []
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 1374 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL BaseModelOutputWithPastAndCrossAttentions []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST hidden_states [UserDefinedClassVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST present_key_value_states [UserDefinedClassVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST all_hidden_states [UserDefinedClassVariable(), TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST all_attentions [UserDefinedClassVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST all_cross_attentions [UserDefinedClassVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST ('last_hidden_state', 'past_key_values', 'hidden_states', 'attentions', 'cross_attentions') [UserDefinedClassVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION_KW 5 [UserDefinedClassVariable(), TensorVariable(), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(NoneType), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [DataClassVariable()]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object forward at 0x7f3f100ba030, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 892>
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST encoder_outputs [DataClassVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE JUMP_FORWARD 194 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST encoder_outputs []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [DataClassVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [DataClassVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST hidden_states [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR model_parallel [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 224 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST labels []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [TensorVariable(), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 264 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST decoder_input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 264 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST decoder_inputs_embeds []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is [ConstantVariable(NoneType), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 264 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR _shift_right [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST labels [UserMethodVariable(<function T5PreTrainedModel._shift_right at 0x7f3ef771cee0>, NNModuleVariable())]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [UserMethodVariable(<function T5PreTrainedModel._shift_right at 0x7f3ef771cee0>, NNModuleVariable()), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object _shift_right at 0x7f3f100b47c0, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 805>
806 0 LOAD_FAST 0 (self)
2 LOAD_ATTR 0 (config)
4 LOAD_ATTR 1 (decoder_start_token_id)
6 STORE_FAST 2 (decoder_start_token_id)
807 8 LOAD_FAST 0 (self)
10 LOAD_ATTR 0 (config)
12 LOAD_ATTR 2 (pad_token_id)
14 STORE_FAST 3 (pad_token_id)
809 16 LOAD_FAST 2 (decoder_start_token_id)
18 LOAD_CONST 0 (None)
20 COMPARE_OP 9 (is not)
22 POP_JUMP_IF_TRUE 32
24 LOAD_GLOBAL 3 (AssertionError)
810 26 LOAD_CONST 1 ('self.model.config.decoder_start_token_id has to be defined. In T5 it is usually set to the pad_token_id. See T5 docs for more information')
809 28 CALL_FUNCTION 1
30 RAISE_VARARGS 1
815 >> 32 LOAD_GLOBAL 4 (is_torch_fx_proxy)
34 LOAD_FAST 1 (input_ids)
36 CALL_FUNCTION 1
38 POP_JUMP_IF_FALSE 98
817 40 LOAD_GLOBAL 5 (torch)
42 LOAD_METHOD 6 (full)
44 LOAD_FAST 1 (input_ids)
46 LOAD_ATTR 7 (shape)
48 LOAD_CONST 0 (None)
50 LOAD_CONST 2 (-1)
52 BUILD_SLICE 2
54 BINARY_SUBSCR
56 LOAD_CONST 3 ((1,))
58 BINARY_ADD
60 LOAD_FAST 2 (decoder_start_token_id)
62 CALL_METHOD 2
64 STORE_FAST 4 (shifted_input_ids)
818 66 LOAD_GLOBAL 5 (torch)
68 LOAD_ATTR 8 (cat)
70 LOAD_FAST 4 (shifted_input_ids)
72 LOAD_FAST 1 (input_ids)
74 LOAD_CONST 4 (Ellipsis)
76 LOAD_CONST 0 (None)
78 LOAD_CONST 2 (-1)
80 BUILD_SLICE 2
82 BUILD_TUPLE 2
84 BINARY_SUBSCR
86 BUILD_LIST 2
88 LOAD_CONST 2 (-1)
90 LOAD_CONST 5 (('dim',))
92 CALL_FUNCTION_KW 2
94 STORE_FAST 4 (shifted_input_ids)
96 JUMP_FORWARD 52 (to 150)
820 >> 98 LOAD_FAST 1 (input_ids)
100 LOAD_METHOD 9 (new_zeros)
102 LOAD_FAST 1 (input_ids)
104 LOAD_ATTR 7 (shape)
106 CALL_METHOD 1
108 STORE_FAST 4 (shifted_input_ids)
821 110 LOAD_FAST 1 (input_ids)
112 LOAD_CONST 4 (Ellipsis)
114 LOAD_CONST 0 (None)
116 LOAD_CONST 2 (-1)
118 BUILD_SLICE 2
120 BUILD_TUPLE 2
122 BINARY_SUBSCR
124 LOAD_METHOD 10 (clone)
126 CALL_METHOD 0
128 LOAD_FAST 4 (shifted_input_ids)
130 LOAD_CONST 4 (Ellipsis)
132 LOAD_CONST 6 (1)
134 LOAD_CONST 0 (None)
136 BUILD_SLICE 2
138 BUILD_TUPLE 2
140 STORE_SUBSCR
822 142 LOAD_FAST 2 (decoder_start_token_id)
144 LOAD_FAST 4 (shifted_input_ids)
146 LOAD_CONST 7 ((Ellipsis, 0))
148 STORE_SUBSCR
824 >> 150 LOAD_FAST 3 (pad_token_id)
152 LOAD_CONST 0 (None)
154 COMPARE_OP 9 (is not)
156 POP_JUMP_IF_TRUE 166
158 LOAD_GLOBAL 3 (AssertionError)
160 LOAD_CONST 8 ('self.model.config.pad_token_id has to be defined.')
162 CALL_FUNCTION 1
164 RAISE_VARARGS 1
826 >> 166 LOAD_FAST 4 (shifted_input_ids)
168 LOAD_METHOD 11 (masked_fill_)
170 LOAD_FAST 4 (shifted_input_ids)
172 LOAD_CONST 9 (-100)
174 COMPARE_OP 2 (==)
176 LOAD_FAST 3 (pad_token_id)
178 CALL_METHOD 2
180 POP_TOP
828 182 LOAD_GLOBAL 5 (torch)
184 LOAD_METHOD 12 (all)
186 LOAD_FAST 4 (shifted_input_ids)
188 LOAD_CONST 10 (0)
190 COMPARE_OP 5 (>=)
192 CALL_METHOD 1
194 LOAD_METHOD 13 (item)
196 CALL_METHOD 0
198 POP_JUMP_IF_TRUE 208
200 LOAD_GLOBAL 3 (AssertionError)
202 LOAD_CONST 11 ('Verify that `shifted_input_ids` has only positive values')
204 CALL_FUNCTION 1
206 RAISE_VARARGS 1
830 >> 208 LOAD_FAST 4 (shifted_input_ids)
210 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR decoder_start_token_id [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST decoder_start_token_id [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST self []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR config [NNModuleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR pad_token_id [HFPretrainedConfigVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST pad_token_id [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST decoder_start_token_id []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 32 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL is_torch_fx_proxy []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids [UserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [UserFunctionVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object is_torch_fx_proxy at 0x7f3f198c6870, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/utils/import_utils.py", line 848>
849 0 LOAD_GLOBAL 0 (is_torch_fx_available)
2 CALL_FUNCTION 0
4 POP_JUMP_IF_FALSE 28
850 6 LOAD_CONST 1 (0)
8 LOAD_CONST 0 (None)
10 IMPORT_NAME 1 (torch.fx)
12 STORE_FAST 1 (torch)
852 14 LOAD_GLOBAL 2 (isinstance)
16 LOAD_FAST 0 (x)
18 LOAD_FAST 1 (torch)
20 LOAD_ATTR 3 (fx)
22 LOAD_ATTR 4 (Proxy)
24 CALL_FUNCTION 2
26 RETURN_VALUE
853 >> 28 LOAD_CONST 2 (False)
30 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL is_torch_fx_available []
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [UserFunctionVariable()]
torchdynamo.symbolic_convert: [DEBUG] INLINING <code object is_torch_fx_available at 0x7f3f198c1190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/utils/import_utils.py", line 359>
360 0 LOAD_GLOBAL 0 (_torch_fx_available)
2 RETURN_VALUE
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL _torch_fx_available []
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object is_torch_fx_available at 0x7f3f198c1190, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/utils/import_utils.py", line 359>
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 28 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE IMPORT_NAME torch.fx [ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST torch [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL isinstance []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST x [BuiltinVariable(isinstance)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST torch [BuiltinVariable(isinstance), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR fx [BuiltinVariable(isinstance), TensorVariable(), TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR Proxy [BuiltinVariable(isinstance), TensorVariable(), TorchVariable(<module 'torch.fx' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/fx/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [BuiltinVariable(isinstance), TensorVariable(), TorchVariable(<class 'torch.fx.proxy.Proxy'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE RETURN_VALUE None [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] DONE INLINING <code object is_torch_fx_proxy at 0x7f3f198c6870, file "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/utils/import_utils.py", line 848>
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_FALSE 98 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR new_zeros [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids [GetAttrVariable(TensorVariable(), new_zeros)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR shape [GetAttrVariable(TensorVariable(), new_zeros), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [GetAttrVariable(TensorVariable(), new_zeros), ShapeVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_FAST shifted_input_ids [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST Ellipsis [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), ConstantVariable(ellipsis)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -1 [TensorVariable(), ConstantVariable(ellipsis), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), ConstantVariable(ellipsis), ConstantVariable(NoneType), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TensorVariable(), ConstantVariable(ellipsis), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE BINARY_SUBSCR None [TensorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR clone [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), clone)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST shifted_input_ids [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST Ellipsis [TensorVariable(), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 1 [TensorVariable(), TensorVariable(), ConstantVariable(ellipsis)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [TensorVariable(), TensorVariable(), ConstantVariable(ellipsis), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_SLICE 2 [TensorVariable(), TensorVariable(), ConstantVariable(ellipsis), ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE BUILD_TUPLE 2 [TensorVariable(), TensorVariable(), ConstantVariable(ellipsis), SliceVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_SUBSCR None [TensorVariable(), TensorVariable(), TupleVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST decoder_start_token_id []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST shifted_input_ids [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST (Ellipsis, 0) [ConstantVariable(int), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE STORE_SUBSCR None [ConstantVariable(int), TensorVariable(), ConstantVariable(tuple)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST pad_token_id []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST None [ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP is not [ConstantVariable(int), ConstantVariable(NoneType)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_JUMP_IF_TRUE 166 [ConstantVariable(bool)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST shifted_input_ids []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR masked_fill_ [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST shifted_input_ids [GetAttrVariable(TensorVariable(), masked_fill_)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST -100 [GetAttrVariable(TensorVariable(), masked_fill_), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP == [GetAttrVariable(TensorVariable(), masked_fill_), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST pad_token_id [GetAttrVariable(TensorVariable(), masked_fill_), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 2 [GetAttrVariable(TensorVariable(), masked_fill_), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE POP_TOP None [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_GLOBAL torch []
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR all [TorchVariable(<module 'torch' from '/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/torch/__init__.py'>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_FAST shifted_input_ids [TorchVariable(<built-in method all of type object at 0x7f40b0102e20>)]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_CONST 0 [TorchVariable(<built-in method all of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE COMPARE_OP >= [TorchVariable(<built-in method all of type object at 0x7f40b0102e20>), TensorVariable(), ConstantVariable(int)]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 1 [TorchVariable(<built-in method all of type object at 0x7f40b0102e20>), TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE LOAD_ATTR item [TensorVariable()]
torchdynamo.symbolic_convert: [DEBUG] TRACE CALL_FUNCTION 0 [GetAttrVariable(TensorVariable(), item)]
torchdynamo.symbolic_convert: [WARNING] Graph break: Tensor.item from user code at File "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 1624, in forward
decoder_input_ids = self._shift_right(labels)
File "/data/home/dberard/miniconda/envs/bench-fast/lib/python3.8/site-packages/transformers/models/t5/modeling_t5.py", line 828, in _shift_right
assert torch.all(shifted_input_ids >= 0).item(), "Verify that `shifted_input_ids` has only positive values"
torchdynamo.output_graph: [INFO] TRACED GRAPH
__compiled_fn_1 <eval_with_key>.3 opcode name target args kwargs
------------- --------------------------------------------------------------------- -------------------------------------------------------------------- ---------------------------------------------------------------------------------- --------------------------------------------------------------
placeholder input_ids input_ids () {}
call_method view view (input_ids, -1, 1024) {}
call_module self_encoder_embed_tokens self_encoder_embed_tokens (view,) {}
call_function ones <built-in method ones of type object at 0x7f40b0102e20> (8, 1024) {}
call_method to to (ones, device(type='cuda', index=0)) {}
call_function getitem <built-in function getitem> (to, (slice(None, None, None), None, None, slice(None, None, None))) {}
call_method to_1 to (getitem,) {'dtype': torch.float32}
call_function sub <built-in function sub> (1.0, to_1) {}
call_function mul <built-in function mul> (sub, -10000.0) {}
call_module self_encoder_dropout self_encoder_dropout (self_encoder_embed_tokens,) {}
call_method to_2 to (self_encoder_dropout, torch.float32) {}
call_method pow_1 pow (to_2, 2) {}
call_method mean mean (pow_1, -1) {'keepdim': True}
call_function add <built-in function add> (mean, 1e-06) {}
call_function rsqrt <built-in method rsqrt of type object at 0x7f40b0102e20> (add,) {}
call_function mul_1 <built-in function mul> (self_encoder_dropout, rsqrt) {}
get_attr self_encoder_block_0__layer_0__layer_norm_weight_2 self_encoder_block_0__layer_0__layer_norm_weight () {}
call_function mul_2 <built-in function mul> (self_encoder_block_0__layer_0__layer_norm_weight_2, mul_1) {}
call_module self_encoder_block_0__layer_0__self_attention_q self_encoder_block_0__layer_0__SelfAttention_q (mul_2,) {}
call_method view_1 view (self_encoder_block_0__layer_0__self_attention_q, 8, -1, 8, 64) {}
call_method transpose transpose (view_1, 1, 2) {}
call_module self_encoder_block_0__layer_0__self_attention_k self_encoder_block_0__layer_0__SelfAttention_k (mul_2,) {}
call_method view_2 view (self_encoder_block_0__layer_0__self_attention_k, 8, -1, 8, 64) {}
call_method transpose_1 transpose (view_2, 1, 2) {}
call_module self_encoder_block_0__layer_0__self_attention_v self_encoder_block_0__layer_0__SelfAttention_v (mul_2,) {}
call_method view_3 view (self_encoder_block_0__layer_0__self_attention_v, 8, -1, 8, 64) {}
call_method transpose_2 transpose (view_3, 1, 2) {}
call_method transpose_3 transpose (transpose_1, 3, 2) {}
call_function matmul <built-in method matmul of type object at 0x7f40b0102e20> (transpose, transpose_3) {}
call_function arange <built-in method arange of type object at 0x7f40b0102e20> (1024,) {'dtype': torch.int64, 'device': device(type='cuda', index=0)}
call_function getitem_1 <built-in function getitem> (arange, (slice(None, None, None), None)) {}
call_function arange_1 <built-in method arange of type object at 0x7f40b0102e20> (1024,) {'dtype': torch.int64, 'device': device(type='cuda', index=0)}
call_function getitem_2 <built-in function getitem> (arange_1, (None, slice(None, None, None))) {}
call_function sub_1 <built-in function sub> (getitem_2, getitem_1) {}
call_function gt <built-in function gt> (sub_1, 0) {}
call_method to_3 to (gt, torch.int64) {}
call_function mul_3 <built-in function mul> (to_3, 16) {}
call_function add_1 <built-in function add> (mul_3, 0) {}
call_function abs_1 <built-in method abs of type object at 0x7f40b0102e20> (sub_1,) {}
call_function lt <built-in function lt> (abs_1, 8) {}
call_method float_1 float (abs_1,) {}
call_function truediv <built-in function truediv> (float_1, 8) {}
call_function log <built-in method log of type object at 0x7f40b0102e20> (truediv,) {}
call_function truediv_1 <built-in function truediv> (log, 2.772588722239781) {}
call_function mul_4 <built-in function mul> (truediv_1, 8) {}
call_method to_4 to (mul_4, torch.int64) {}
call_function add_2 <built-in function add> (8, to_4) {}
call_function full_like <built-in method full_like of type object at 0x7f40b0102e20> (add_2, 15) {}
call_function min_1 <built-in method min of type object at 0x7f40b0102e20> (add_2, full_like) {}
call_function where <built-in method where of type object at 0x7f40b0102e20> (lt, abs_1, min_1) {}
call_function iadd <built-in function iadd> (add_1, where) {}
call_module self_encoder_block_0__layer_0__self_attention_relative_attention_bias self_encoder_block_0__layer_0__SelfAttention_relative_attention_bias (iadd,) {}
call_method permute permute (self_encoder_block_0__layer_0__self_attention_relative_attention_bias, [2, 0, 1]) {}
call_method unsqueeze unsqueeze (permute, 0) {}
call_function add_3 <built-in function add> (unsqueeze, mul) {}
call_function iadd_1 <built-in function iadd> (matmul, add_3) {}
call_method float_2 float (iadd_1,) {}
call_function softmax <function softmax at 0x7f3f1bf53e50> (float_2,) {'dim': -1}
call_method type_as type_as (softmax, iadd_1) {}
call_function lowmem_dropout_5 <function lowmem_dropout at 0x7f3ef6b84d30> (type_as,) {'p': 0.1, 'training': True}
call_function matmul_1 <built-in method matmul of type object at 0x7f40b0102e20> (lowmem_dropout_5, transpose_2) {}
call_method transpose_4 transpose (matmul_1, 1, 2) {}
call_method contiguous contiguous (transpose_4,) {}
call_method view_4 view (contiguous, 8, -1, 512) {}
call_module self_encoder_block_0__layer_0__self_attention_o self_encoder_block_0__layer_0__SelfAttention_o (view_4,) {}
call_module self_encoder_block_0__layer_0__dropout self_encoder_block_0__layer_0__dropout (self_encoder_block_0__layer_0__self_attention_o,) {}
call_function add_4 <built-in function add> (self_encoder_dropout, self_encoder_block_0__layer_0__dropout) {}
call_method to_5 to (add_4, torch.float32) {}
call_method pow_2 pow (to_5, 2) {}
call_method mean_1 mean (pow_2, -1) {'keepdim': True}
call_function add_5 <built-in function add> (mean_1, 1e-06) {}
call_function rsqrt_1 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_5,) {}
call_function mul_5 <built-in function mul> (add_4, rsqrt_1) {}
get_attr self_encoder_block_0__layer_1__layer_norm_weight_2 self_encoder_block_0__layer_1__layer_norm_weight () {}
call_function mul_6 <built-in function mul> (self_encoder_block_0__layer_1__layer_norm_weight_2, mul_5) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_wi self_encoder_block_0__layer__1__DenseReluDense_wi (mul_6,) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_act self_encoder_block_0__layer__1__DenseReluDense_act (self_encoder_block_0__layer__1__dense_relu_dense_wi,) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_dropout self_encoder_block_0__layer__1__DenseReluDense_dropout (self_encoder_block_0__layer__1__dense_relu_dense_act,) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_wo self_encoder_block_0__layer__1__DenseReluDense_wo (self_encoder_block_0__layer__1__dense_relu_dense_dropout,) {}
call_module self_encoder_block_0__layer__1__dropout self_encoder_block_0__layer__1__dropout (self_encoder_block_0__layer__1__dense_relu_dense_wo,) {}
call_function add_6 <built-in function add> (add_4, self_encoder_block_0__layer__1__dropout) {}
call_method to_6 to (add_6, torch.float32) {}
call_method pow_3 pow (to_6, 2) {}
call_method mean_2 mean (pow_3, -1) {'keepdim': True}
call_function add_7 <built-in function add> (mean_2, 1e-06) {}
call_function rsqrt_2 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_7,) {}
call_function mul_7 <built-in function mul> (add_6, rsqrt_2) {}
get_attr self_encoder_block_1__layer_0__layer_norm_weight_2 self_encoder_block_1__layer_0__layer_norm_weight () {}
call_function mul_8 <built-in function mul> (self_encoder_block_1__layer_0__layer_norm_weight_2, mul_7) {}
call_module self_encoder_block_1__layer_0__self_attention_q self_encoder_block_1__layer_0__SelfAttention_q (mul_8,) {}
call_method view_5 view (self_encoder_block_1__layer_0__self_attention_q, 8, -1, 8, 64) {}
call_method transpose_5 transpose (view_5, 1, 2) {}
call_module self_encoder_block_1__layer_0__self_attention_k self_encoder_block_1__layer_0__SelfAttention_k (mul_8,) {}
call_method view_6 view (self_encoder_block_1__layer_0__self_attention_k, 8, -1, 8, 64) {}
call_method transpose_6 transpose (view_6, 1, 2) {}
call_module self_encoder_block_1__layer_0__self_attention_v self_encoder_block_1__layer_0__SelfAttention_v (mul_8,) {}
call_method view_7 view (self_encoder_block_1__layer_0__self_attention_v, 8, -1, 8, 64) {}
call_method transpose_7 transpose (view_7, 1, 2) {}
call_method transpose_8 transpose (transpose_6, 3, 2) {}
call_function matmul_2 <built-in method matmul of type object at 0x7f40b0102e20> (transpose_5, transpose_8) {}
call_function iadd_2 <built-in function iadd> (matmul_2, add_3) {}
call_method float_3 float (iadd_2,) {}
call_function softmax_1 <function softmax at 0x7f3f1bf53e50> (float_3,) {'dim': -1}
call_method type_as_1 type_as (softmax_1, iadd_2) {}
call_function lowmem_dropout_4 <function lowmem_dropout at 0x7f3ef6b84d30> (type_as_1,) {'p': 0.1, 'training': True}
call_function matmul_3 <built-in method matmul of type object at 0x7f40b0102e20> (lowmem_dropout_4, transpose_7) {}
call_method transpose_9 transpose (matmul_3, 1, 2) {}
call_method contiguous_1 contiguous (transpose_9,) {}
call_method view_8 view (contiguous_1, 8, -1, 512) {}
call_module self_encoder_block_1__layer_0__self_attention_o self_encoder_block_1__layer_0__SelfAttention_o (view_8,) {}
call_module self_encoder_block_1__layer_0__dropout self_encoder_block_1__layer_0__dropout (self_encoder_block_1__layer_0__self_attention_o,) {}
call_function add_8 <built-in function add> (add_6, self_encoder_block_1__layer_0__dropout) {}
call_method to_7 to (add_8, torch.float32) {}
call_method pow_4 pow (to_7, 2) {}
call_method mean_3 mean (pow_4, -1) {'keepdim': True}
call_function add_9 <built-in function add> (mean_3, 1e-06) {}
call_function rsqrt_3 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_9,) {}
call_function mul_9 <built-in function mul> (add_8, rsqrt_3) {}
get_attr self_encoder_block_1__layer_1__layer_norm_weight_2 self_encoder_block_1__layer_1__layer_norm_weight () {}
call_function mul_10 <built-in function mul> (self_encoder_block_1__layer_1__layer_norm_weight_2, mul_9) {}
call_module self_encoder_block_1__layer__1__dense_relu_dense_wi self_encoder_block_1__layer__1__DenseReluDense_wi (mul_10,) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_act_1 self_encoder_block_0__layer__1__DenseReluDense_act (self_encoder_block_1__layer__1__dense_relu_dense_wi,) {}
call_module self_encoder_block_1__layer__1__dense_relu_dense_dropout self_encoder_block_1__layer__1__DenseReluDense_dropout (self_encoder_block_0__layer__1__dense_relu_dense_act_1,) {}
call_module self_encoder_block_1__layer__1__dense_relu_dense_wo self_encoder_block_1__layer__1__DenseReluDense_wo (self_encoder_block_1__layer__1__dense_relu_dense_dropout,) {}
call_module self_encoder_block_1__layer__1__dropout self_encoder_block_1__layer__1__dropout (self_encoder_block_1__layer__1__dense_relu_dense_wo,) {}
call_function add_10 <built-in function add> (add_8, self_encoder_block_1__layer__1__dropout) {}
call_method to_8 to (add_10, torch.float32) {}
call_method pow_5 pow (to_8, 2) {}
call_method mean_4 mean (pow_5, -1) {'keepdim': True}
call_function add_11 <built-in function add> (mean_4, 1e-06) {}
call_function rsqrt_4 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_11,) {}
call_function mul_11 <built-in function mul> (add_10, rsqrt_4) {}
get_attr self_encoder_block_2__layer_0__layer_norm_weight_2 self_encoder_block_2__layer_0__layer_norm_weight () {}
call_function mul_12 <built-in function mul> (self_encoder_block_2__layer_0__layer_norm_weight_2, mul_11) {}
call_module self_encoder_block_2__layer_0__self_attention_q self_encoder_block_2__layer_0__SelfAttention_q (mul_12,) {}
call_method view_9 view (self_encoder_block_2__layer_0__self_attention_q, 8, -1, 8, 64) {}
call_method transpose_10 transpose (view_9, 1, 2) {}
call_module self_encoder_block_2__layer_0__self_attention_k self_encoder_block_2__layer_0__SelfAttention_k (mul_12,) {}
call_method view_10 view (self_encoder_block_2__layer_0__self_attention_k, 8, -1, 8, 64) {}
call_method transpose_11 transpose (view_10, 1, 2) {}
call_module self_encoder_block_2__layer_0__self_attention_v self_encoder_block_2__layer_0__SelfAttention_v (mul_12,) {}
call_method view_11 view (self_encoder_block_2__layer_0__self_attention_v, 8, -1, 8, 64) {}
call_method transpose_12 transpose (view_11, 1, 2) {}
call_method transpose_13 transpose (transpose_11, 3, 2) {}
call_function matmul_4 <built-in method matmul of type object at 0x7f40b0102e20> (transpose_10, transpose_13) {}
call_function iadd_3 <built-in function iadd> (matmul_4, add_3) {}
call_method float_4 float (iadd_3,) {}
call_function softmax_2 <function softmax at 0x7f3f1bf53e50> (float_4,) {'dim': -1}
call_method type_as_2 type_as (softmax_2, iadd_3) {}
call_function lowmem_dropout_3 <function lowmem_dropout at 0x7f3ef6b84d30> (type_as_2,) {'p': 0.1, 'training': True}
call_function matmul_5 <built-in method matmul of type object at 0x7f40b0102e20> (lowmem_dropout_3, transpose_12) {}
call_method transpose_14 transpose (matmul_5, 1, 2) {}
call_method contiguous_2 contiguous (transpose_14,) {}
call_method view_12 view (contiguous_2, 8, -1, 512) {}
call_module self_encoder_block_2__layer_0__self_attention_o self_encoder_block_2__layer_0__SelfAttention_o (view_12,) {}
call_module self_encoder_block_2__layer_0__dropout self_encoder_block_2__layer_0__dropout (self_encoder_block_2__layer_0__self_attention_o,) {}
call_function add_12 <built-in function add> (add_10, self_encoder_block_2__layer_0__dropout) {}
call_method to_9 to (add_12, torch.float32) {}
call_method pow_6 pow (to_9, 2) {}
call_method mean_5 mean (pow_6, -1) {'keepdim': True}
call_function add_13 <built-in function add> (mean_5, 1e-06) {}
call_function rsqrt_5 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_13,) {}
call_function mul_13 <built-in function mul> (add_12, rsqrt_5) {}
get_attr self_encoder_block_2__layer_1__layer_norm_weight_2 self_encoder_block_2__layer_1__layer_norm_weight () {}
call_function mul_14 <built-in function mul> (self_encoder_block_2__layer_1__layer_norm_weight_2, mul_13) {}
call_module self_encoder_block_2__layer__1__dense_relu_dense_wi self_encoder_block_2__layer__1__DenseReluDense_wi (mul_14,) {}
call_module self_encoder_block_0__layer__1__dense_relu_dense_act_2 self_encoder_block_0__layer__1__DenseReluDense_act (self_encoder_block_2__layer__1__dense_relu_dense_wi,) {}
call_module self_encoder_block_2__layer__1__dense_relu_dense_dropout self_encoder_block_2__layer__1__DenseReluDense_dropout (self_encoder_block_0__layer__1__dense_relu_dense_act_2,) {}
call_module self_encoder_block_2__layer__1__dense_relu_dense_wo self_encoder_block_2__layer__1__DenseReluDense_wo (self_encoder_block_2__layer__1__dense_relu_dense_dropout,) {}
call_module self_encoder_block_2__layer__1__dropout self_encoder_block_2__layer__1__dropout (self_encoder_block_2__layer__1__dense_relu_dense_wo,) {}
call_function add_14 <built-in function add> (add_12, self_encoder_block_2__layer__1__dropout) {}
call_method to_10 to (add_14, torch.float32) {}
call_method pow_7 pow (to_10, 2) {}
call_method mean_6 mean (pow_7, -1) {'keepdim': True}
call_function add_15 <built-in function add> (mean_6, 1e-06) {}
call_function rsqrt_6 <built-in method rsqrt of type object at 0x7f40b0102e20> (add_15,) {}
call_function mul_15 <built-in function mul> (add_14, rsqrt_6) {}
get_attr self_encoder_block_3__layer_0__layer_norm_weight_2 self_encoder_block_3__layer_0__layer_norm_weight () {}
call_function mul_16 <built-in function mul> (self_encoder_block_3__layer_0__layer_norm_weight_2, mul_15) {}
call_module self_encoder_block_3__layer_0__self_attention_q self_encoder_block_3__layer_0__SelfAttention_q (mul_16,) {}
call_method view_13 view (self_encoder_block_3__layer_0__self_attention_q, 8, -1, 8, 64) {}
call_method transpose_15 transpose (view_13, 1, 2) {}
call_module self_encoder_block_3__layer_0__self_attention_k self_encoder_block_3__layer_0__SelfAttention_k (mul_16,) {}
call_method view_14 view (self_encoder_block_3__layer_0__self_attention_k, 8, -1, 8, 64) {}
call_method transpose_16 transpose (view_14, 1, 2) {}
call_module self_encoder_block_3__layer_0__self_attention_v self_encoder_block_3__layer_0__SelfAttention_v (mul_16,) {}
call_method view_15 view (self_encoder_block_3__layer_0__self_attention_v, 8, -1, 8, 64) {}
call_method transpose_17 transpose (view_15, 1, 2) {}
call_method transpose_18 transpose (transpose_16, 3, 2) {}
call_function matmul_6 <built-in method matmul of type object at 0x7f40b0102e20> (transpose_15, transpose_18) {}
call_function iadd_4 <built-in function iadd> (matmul_6, add_3) {}
call_method float_5 float (iadd_4,) {}
call_function softmax_3 <function softmax at 0x7f3f1bf53e50> (float_5,) {'dim': -1}
call_method type_as_3 type_as (softmax_3, iadd_4) {}
call_function lowmem_dropout_2 <function lowmem_dropout at 0x7f3ef6b84d30> (type_as_3,) {'p': 0.1, 'training': True}
call_function matmul_7 <built-in method matmul of type object at 0x7f40b0102e20> (lowmem_dropout_2, transpose_17) {}
call_method transpose_19 transpose (matmul_7, 1, 2) {}
call_method contiguous_3 contiguous (transpose_19,) {}
call_method view_16 view (contiguous_3, 8, -1, 512) {}
call_module self_encoder_block_3__layer_0__self_attention_o self_encoder_block_3__layer_0__SelfAttention_o (view_16,) {}
call_module self_encoder_block_3__layer_0__dropout self_encoder_block_3__layer_0__dropout (self_encoder_block_3__layer_0__self_attention_o,) {}
call_func
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment