Config for https://github.com/pytorch/torchtune/pull/2721

gistfile1.txt

batch_size: 16
batch_size_val: 8
checkpointer:
  _component_: torchtune.training.FullModelHFCheckpointer
  checkpoint_dir: /tmp/Llama-3.2-3B-Instruct/
  checkpoint_files:
  - model-00001-of-00002.safetensors
  - model-00002-of-00002.safetensors
  model_type: LLAMA3_2
  output_dir: /home/andrewor/local/logs/tune/Llama3.2-3B_qat
  recipe_checkpoint: null
clip_grad_norm: null
compile: false
dataset:
  _component_: torchtune.datasets.alpaca_cleaned_dataset
  packed: false
  split: train[:95%]
dataset_val:
  _component_: torchtune.datasets.alpaca_cleaned_dataset
  split: train[95%:]
device: cuda
dtype: bf16
enable_activation_checkpointing: true
enable_activation_offloading: false
epochs: 1
gradient_accumulation_steps: 8
log_every_n_steps: 1
log_peak_memory_stats: true
loss:
  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
max_steps_per_epoch: null
metric_logger:
  _component_: torchtune.training.metric_logging.DiskLogger
  log_dir: /home/andrewor/local/logs/tune/Llama3.2-3B_qat/metrics
model:
  _component_: torchtune.models.llama3_2.llama3_2_3b
optimizer:
  _component_: torch.optim.AdamW
  fused: true
  lr: 2.0e-05
optimizer_in_bwd: false
output_dir: /home/andrewor/local/logs/tune/Llama3.2-3B_qat/metrics
profiler:
  _component_: torchtune.training.setup_torch_profiler
  active_steps: 2
  cpu: true
  cuda: true
  enabled: false
  num_cycles: 1
  output_dir: /home/andrewor/local/logs/tune/Llama3.2-3B_qat/metrics/profiling_outputs
  profile_memory: false
  record_shapes: true
  wait_steps: 5
  warmup_steps: 3
  with_flops: false
  with_stack: false
quantizer:
  _component_: torchtune.training.quantization.Int8DynActInt4WeightQATQuantizer
  groupsize: 32
resume_from_checkpoint: false
run_val_every_n_steps: null
seed: null
shuffle: true
tokenizer:
  _component_: torchtune.models.llama3.llama3_tokenizer
  max_seq_len: null
  path: /tmp/Llama-3.2-3B-Instruct/original/tokenizer.model