model:
  _component_: torchtune.models.llama3_2_vision.llama3_2_vision_11b
  decoder_trainable: false
  encoder_trainable: true
  fusion_trainable: true
  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
checkpointer:
  _component_: torchtune.training.FullModelMetaCheckpointer
  checkpoint_dir: /tmp/Llama-3.2-11B-Vision-Instruct/original/
  checkpoint_files:
  - consolidated.pth
  recipe_checkpoint: null
  output_dir: /tmp/Llama-3.2-11B-Vision-Instruct/
  model_type: LLAMA3_VISION
resume_from_checkpoint: false
dataset:
  _component_: data.chart_dataset
  source: jrc/data-viz-qa
  split: train
seed: 42
shuffle: true
collate_fn: torchtune.data.padded_collate_tiled_images_and_mask
epochs: 2
max_steps_per_epoch: null
batch_size: 8
gradient_accumulation_steps: 4
optimizer:
  _component_: torch.optim.AdamW
  lr: 2.0e-05
  fused: true
loss:
  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss
clip_grad_norm: 1.0
compile: false
device: cuda
enable_activation_checkpointing: true
custom_sharded_layers: []
dtype: bf16
output_dir: /tmp/full-llama3.2-vision--finetune
metric_logger:
  _component_: torchtune.training.metric_logging.WandBLogger
  project: plot-huh
  name: dataviz-qa-full
log_every_n_steps: 1
log_peak_memory_stats: true