model:
  _component_: torchtune.models.llama3_2_vision.qlora_llama3_2_vision_11b
  decoder_trainable: frozen
  encoder_trainable: lora
  fusion_trainable: lora
  lora_attn_modules:
  - q_proj
  - v_proj
  - output_proj
  apply_lora_to_mlp: true
  apply_lora_to_output: false
  lora_rank: 8
  lora_alpha: 16
  lora_dropout: 0.0
  image_size: 560
tokenizer:
  _component_: torchtune.models.llama3_2_vision.llama3_2_vision_transform
  path: /tmp/Llama-3.2-11B-Vision-Instruct/original/tokenizer.model
  image_size: 560
  max_seq_len: 8192
checkpointer:
  _component_: torchtune.training.FullModelHFCheckpointer
  checkpoint_dir: /tmp/Llama-3.2-11B-Vision-Instruct/
  checkpoint_files:
    filename_format: model-{}-of-{}.safetensors
    max_filename: '00005'
  recipe_checkpoint: null
  output_dir: /tmp/Llama-3.2-11B-Vision-Instruct/
  model_type: LLAMA3_VISION
resume_from_checkpoint: false
save_adapter_weights_only: false
dataset:
  _component_: torchtune.datasets.multimodal.the_cauldron_dataset
  subset: diagram_image_to_text
seed: null
shuffle: true
collate_fn: torchtune.data.padded_collate_tiled_images_and_mask
epochs: 3
max_steps_per_epoch: null
batch_size: 2
gradient_accumulation_steps: 8
optimizer:
  _component_: torch.optim.AdamW
  fused: true
  weight_decay: 0.01
  lr: 0.0001
lr_scheduler:
  _component_: torchtune.training.lr_schedulers.get_cosine_schedule_with_warmup
  num_warmup_steps: 100
loss:
  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
clip_grad_norm: 1.0
compile: false
device: cuda
enable_activation_checkpointing: true
dtype: bf16
output_dir: /tmp/qlora-llama3.2-vision-finetune
metric_logger:
  _component_: torchtune.training.metric_logging.WandBLogger
  project: llama-3.2-vlm-torchtune
log_every_n_steps: 1
log_peak_memory_stats: true
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: 3
  active_steps: 2
  num_cycles: 1