pashu123
/ test_model.py
Created
February 14, 2023 07:31
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| from transformers import AutoTokenizer, AutoModelForSeq2SeqLM | |
| import torch | |
| from torch.fx.experimental.proxy_tensor import make_fx | |
| from torch._decomp import get_decompositions | |
| import tempfile | |
| import torch_mlir | |
| def prepare_sentence_tokens(hf_model: str): | |
| tokenizer = AutoTokenizer.from_pretrained(hf_model) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Package Version Editable project location | |
| ---------------------------- ------------------------ ------------------------------ | |
| absl-py 1.2.0 | |
| aiofiles 22.1.0 | |
| aiohttp 3.8.3 | |
| aiosignal 1.2.0 | |
| albumentations 0.4.3 | |
| altair 4.2.0 | |
| antlr4-python3-runtime 4.8 | |
| anyio 3.6.2 |
pashu123
/ linalg_el.mlir
Created
February 14, 2023 17:18
This file has been truncated, but you can view the full file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #map = affine_map<(d0) -> (0)> | |
| #map1 = affine_map<(d0) -> (d0)> | |
| #map2 = affine_map<(d0) -> ()> | |
| #map3 = affine_map<(d0, d1) -> (d0, 0)> | |
| #map4 = affine_map<(d0, d1) -> (0, d1)> | |
| #map5 = affine_map<(d0, d1) -> (d0, d1)> | |
| #map6 = affine_map<(d0, d1) -> (d1, d0)> | |
| #map7 = affine_map<(d0, d1) -> (d1)> | |
| #map8 = affine_map<(d0, d1, d2, d3) -> (d1)> | |
| #map9 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)> |
pashu123
/ torch_el.mlir
Created
February 14, 2023 17:19
This file has been truncated, but you can view the full file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| module attributes {torch.debug_module_name = "_lambda"} { | |
| func.func @forward(%arg0: !torch.vtensor<[2,4,96,96],f16>, %arg1: !torch.vtensor<[1],f32>, %arg2: !torch.vtensor<[4,77,1024],f16>, %arg3: !torch.vtensor<[],f32>) -> !torch.vtensor<[2,96,96],f16> { | |
| %int1 = torch.constant.int 1 | |
| %int0 = torch.constant.int 0 | |
| %false = torch.constant.bool false | |
| %int4 = torch.constant.int 4 | |
| %float0.000000e00 = torch.constant.float 0.000000e+00 | |
| %int5 = torch.constant.int 5 | |
| %none = torch.constant.none | |
| %int320 = torch.constant.int 320 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Args: {'output_path': '.', 'quant_device': 'cuda', 'batch_size': 2, 'bit_width': 8, 'conv_group_size': 16, 'linear_group_size': 16, 'layer_type': 'conv', 'weight_quant': 'group_scale_param', 'input_quant': 'group_dynamic_sym', 'float16': True, 'inline_ts_graph': False, 'onnx_export': False, 'make_fx': True, 'quantize': True} | |
| Moving model to cuda... | |
| Run model quantization... | |
| Quantize conv2d: conv | |
| Tracing to FX functional representation... | |
| FX graph model... | |
| graph(): | |
| %arg0_1 : [#users=2] = placeholder[target=arg0_1] | |
| %_tensor_constant0 : [#users=1] = get_attr[target=_tensor_constant0] | |
| %lift_fresh_copy : [#users=0] = call_function[target=torch.ops.aten.lift_fresh_copy](args = (%_tensor_constant0,), kwargs = {}) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| /home/prashant/llama/llama/model.py:228: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! | |
| start_pos = int(start_pos_tensor) | |
| /home/prashant/llama/llama/model.py:232: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! | |
| if seqlen > 1: | |
| /home/prashant/llama/llama/model.py:58: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs! | |
| assert freqs_cis.shape == (x.shape[1], x.shape[-1]) |
pashu123
/ Transformer.mlir
Created
February 28, 2023 17:48
This file has been truncated, but you can view the full file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #loc = loc(unknown) | |
| module attributes {torch.debug_module_name = "Transformer"} { | |
| func.func private @__torch__.torch.nn.modules.sparse.Embedding.forward(%arg0: !torch.nn.Module<"__torch__.torch.nn.modules.sparse.Embedding"> loc(unknown), %arg1: !torch.tensor loc(unknown)) -> !torch.tensor { | |
| %185 = torch.tensor_static_info_cast %arg1 : !torch.tensor to !torch.tensor<[1,1],si64> loc(#loc) | |
| %186 = torch.prim.GetAttr %arg0["weight"] : !torch.nn.Module<"__torch__.torch.nn.modules.sparse.Embedding"> -> !torch.tensor loc(#loc) | |
| %int-1 = torch.constant.int -1 loc(#loc1) | |
| %false = torch.constant.bool false loc(#loc1) | |
| %false_0 = torch.constant.bool false loc(#loc1) | |
| %187 = torch.aten.embedding %186, %185, %int-1, %false, %false_0 : !torch.tensor, !torch.tensor<[1,1],si64>, !torch.int, !torch.bool, !torch.bool -> !torch.tensor<[1,1,512],f32> loc(#loc1) | |
| %188 = torch.tensor_static_info_cast %187 : !torch.tensor<[1,1,512],f32> to !torch.tensor loc(#loc1) |
pashu123
/ llama_linalg.mlir
Created
March 3, 2023 15:38
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| func.func @forward(%arg0: tensor<512xf32>, %arg1: tensor<512xf32>, %arg2: tensor<512xf32>, %arg3: tensor<512xf32>, %arg4: tensor<512xf32>, %arg5: tensor<512xf32>, %arg6: tensor<512xf32>, %arg7: tensor<512xf32>, %arg8: tensor<512xf32>, %arg9: tensor<512xf32>, %arg10: tensor<512xf32>, %arg11: tensor<512xf32>, %arg12: tensor<512xf32>, %arg13: tensor<512xf32>, %arg14: tensor<512xf32>, %arg15: tensor<512xf32>, %arg16: tensor<512xf32>, %arg17: tensor<1x512xf32>, %arg18: tensor<512x512xf32>, %arg19: tensor<512x512xf32>, %arg20: tensor<512x512xf32>, %arg21: tensor<512x512xf32>, %arg22: tensor<1536x512xf32>, %arg23: tensor<1536x512xf32>, %arg24: tensor<512x1536xf32>, %arg25: tensor<512x512xf32>, %arg26: tensor<512x512xf32>, %arg27: tensor<512x512xf32>, %arg28: tensor<512x512xf32>, %arg29: tensor<1536x512xf32>, %arg30: tensor<1536x512xf32>, %arg31: tensor<512x1536xf32>, %arg32: tensor<512x512xf32>, %arg33: tensor<512x512xf32>, %arg34: tensor<512x512xf32>, %arg35: tensor<512x512xf32>, %arg36: tensor<1536x512xf32>, %arg3 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| from iree import runtime as ireert | |
| from iree.compiler import compile_str | |
| import numpy as np | |
| LINALG_IR = ''' | |
| #map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)> | |
| module { | |
| func.func @forward(%arg0: tensor<1x6x32x64x2xf32>) -> tensor<1x6x32x64xcomplex<f32>> { | |
| %c1 = arith.constant 1 : index |
pashu123
/ int8_elided.mlir
Created
March 20, 2023 15:58
This file has been truncated, but you can view the full file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #map = affine_map<(d0, d1, d2, d3) -> (0, d1, d2, d3)> | |
| #map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)> | |
| #map2 = affine_map<(d0) -> (0)> | |
| #map3 = affine_map<(d0) -> (d0)> | |
| #map4 = affine_map<(d0) -> ()> | |
| #map5 = affine_map<(d0, d1) -> (d0, 0)> | |
| #map6 = affine_map<(d0, d1) -> (0, d1)> | |
| #map7 = affine_map<(d0, d1) -> (d0, d1)> | |
| #map8 = affine_map<(d0, d1) -> (d1, d0)> | |
| #map9 = affine_map<(d0, d1) -> (d1)> |