output_dir: checkpoints/qwen2_5_3B/lora
model:
  _component_: torchtune.models.qwen2_5.lora_qwen2_5_3b
  lora_attn_modules:
  - q_proj
  - v_proj
  - output_proj
  apply_lora_to_mlp: true
  lora_rank: 8
  lora_alpha: 16
  lora_dropout: 0.0
tokenizer:
  _component_: torchtune.models.qwen2_5.qwen2_5_tokenizer
  path: ./Qwen2_5-3B-Instruct/vocab.json
  merges_file: ./Qwen2_5-3B-Instruct/merges.txt
  max_seq_len: null
checkpointer:
  _component_: torchtune.training.FullModelHFCheckpointer
  checkpoint_dir: ./Qwen2_5-3B-Instruct
  checkpoint_files:
  - model-00001-of-00002.safetensors
  - model-00002-of-00002.safetensors
  recipe_checkpoint: null
  output_dir: ${output_dir}
  model_type: QWEN2
resume_from_checkpoint: false
dataset:
  _component_: torchtune.datasets.chat_dataset
  source: json
  data_files: ./rankwogpt-data.json
  conversation_column: conversations
  conversation_style: sharegpt
  train_on_input: false
  split: train
  packed: false
seed: null
shuffle: true
batch_size: 2
optimizer:
  _component_: torch.optim.AdamW
  fused: true
  weight_decay: 0.01
  lr: 0.0003
lr_scheduler:
  _component_: torchtune.training.lr_schedulers.get_cosine_schedule_with_warmup
  num_warmup_steps: 100
loss:
  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
epochs: 1
max_steps_per_epoch: null
gradient_accumulation_steps: 8
compile: false
metric_logger:
  _component_: torchtune.training.metric_logging.WandBLogger
  project: torchtune
log_every_n_steps: 1
log_peak_memory_stats: false
device: cuda
dtype: bf16
enable_activation_checkpointing: false
enable_activation_offloading: false
profiler:
  _component_: torchtune.training.setup_torch_profiler
  enabled: false
  output_dir: ${output_dir}/profiling_outputs
  cpu: true
  cuda: true
  profile_memory: false
  with_stack: false
  record_shapes: true
  with_flops: false
  wait_steps: 5
  warmup_steps: 5
  active_steps: 2
  num_cycles: 1