model:
  _component_: torchtune.models.llama2.qlora_llama2_7b
  lora_attn_modules:
  - q_proj
  - v_proj
  - k_proj
  - output_proj
  apply_lora_to_mlp: true
  apply_lora_to_output: false
  lora_rank: 8
  lora_alpha: 16
tokenizer:
  _component_: torchtune.models.llama2.llama2_tokenizer
  path: ./checkpoint/tokenizer.model
checkpointer:
  _component_: torchtune.utils.FullModelHFCheckpointer
  checkpoint_dir: ./checkpoint
  checkpoint_files:
  - pytorch_model-00001-of-00002.bin
  - pytorch_model-00002-of-00002.bin
  adapter_checkpoint: null
  recipe_checkpoint: null
  output_dir: ./checkpoint
  model_type: LLAMA2
resume_from_checkpoint: false
dataset:
  _component_: torchtune.datasets.alpaca_cleaned_dataset
  train_on_input: true
seed: null
shuffle: true
batch_size: 2
optimizer:
  _component_: torch.optim.AdamW
  weight_decay: 0.01
  lr: 0.0003
lr_scheduler:
  _component_: torchtune.modules.get_cosine_schedule_with_warmup
  num_warmup_steps: 100
loss:
  _component_: torch.nn.CrossEntropyLoss
epochs: 1
max_steps_per_epoch: null
gradient_accumulation_steps: 16
compile: false
output_dir: /tmp/qlora_finetune_output/
metric_logger:
  _component_: torchtune.utils.metric_logging.WandBLogger
  log_dir: ${output_dir}
log_every_n_steps: 1
device: cuda
dtype: bf16
enable_activation_checkpointing: true
profiler:
  _component_: torchtune.utils.profiler
  enabled: false
  output_dir: ${output_dir}/torchtune_perf_tracing.json