run_id: 0906_bestvla_retrain_lr_v2
run_root_dir: ./results/Checkpoints
seed: 42
trackers:
- jsonl
- wandb
wandb_entity: jinhuiye
wandb_project: InternVLA
is_debug: false
framework:
  framework_py: InternVLA-M1
  qwenvl:
    base_vlm: InternRobotics/InternVLA-M1
    attn_implementation: flash_attention_2
    vl_hidden_dim: 2048
  dino:
    dino_backbone: dinov2_vits14
  layer_qformer:
    qformer_end_layer: 37
    qformer_start_layer: 36
    num_query_tokens: 64
    input_dim: 2048
    ouptput_dim: 768
    grad_scale: 0.5
  action_model:
    action_model_type: DiT-B
    action_hidden_dim: 768
    action_dim: 7
    use_ema: false
    future_action_window_size: 15
    past_action_window_size: 0
    repeated_diffusion_steps: 8
  reduce_in_full_precision: true
datasets:
  vlm_data:
    dataformat: llava_json
    dataset_use: ao_droid_data,ao_droid_molmo_sam2,ao_hoi4d_data,ao_maniskills,ao_hoi4d_frame_data,pixmo_point,refspatial_sim%10,xudong_spatial_interact%10,xudong_invalid_task%10,xudong_task_onlyaction%10,xudong_task_cot_cap_resp_act%10,gsys2_14kv2_gd_coco_rule%10,gsys2_14kv2_obj_attr%10,gsys2_14kv2_obj_nearby%10,gsys2_14kv2_obj_senmatic%10,gsys2_14kv2_action_plan%10,asv2_conversation_en,asv2_detailed_description_en,asv2_region_captioning_en,coco_internvl_longcap_en,coco_karpathy_train_567_en,coco_neg_gpt4o_en,coco_poetry_zh,coco_rem_en_zh,cocorem_exist_yorn_en,cocotextv2_en,cocotextv2_gpt4o_en,okvqa_en,refcoco_grounding_aug_en,tallyqa_coco_en,toloka_grounding_aug_en,vqav2_en,vsr_en,robo_point_refobj,robo_point_refregion,roborefit,ao_droid_data,ao_droid_molmo_sam2,ao_hoi4d_data,ao_hoi4d_frame_data,ao_maniskills,molmo_traj_auxiliary_trace,molmo_traj_bridge_dataset,molmo_traj_bc_z,molmo_traj_fractal
    eval_dataset: aokvqa_cauldron_llava_format
    data_flatten: false
    base_interval: 2
    max_pixels: 12845056
    min_pixels: 3136
    model_max_length: 2048
    model_type: qwen2.5vl
    per_device_batch_size: 2
  vla_data:
    dataset_py: rlds_datasets
    data_root_dir: playground/Datasets/OXE_openvla
    data_mix: bridge_rt_1
    CoT_prompt: Your task is {instruction}. To identify the key objects for your task.
      Locate their bounding boxes in [x1,y1,x2,y2] format.
    CoT_answer: bbox
    default_image_resolution:
    - 3
    - 224
    - 224
    shuffle_buffer_size: 250000
    image_aug: true
    per_device_batch_size: 16
    load_all_data_for_training: true
trainer:
  epochs: 100
  max_train_steps: 100000
  num_warmup_steps: 5000
  save_interval: 5000
  eval_interval: 100
  learning_rate:
    base: 4.0e-05
    qwen_vl_interface: 1.0e-05
    action_model: 0.0001
  lr_scheduler_type: cosine_with_min_lr
  scheduler_specific_kwargs:
    min_lr: 5.0e-07
  freeze_modules: null
  loss_scale:
    vla: 1.0
    vlm: 0.1
  max_grad_norm: 1.0
  warmup_ratio: 0.1
  weight_decay: 0.0
  logging_frequency: 10
  gradient_clipping: 1.0
  gradient_accumulation_steps: 1
  optimizer:
    name: AdamW
    betas:
    - 0.9
    - 0.95
    eps: 1.0e-08
    weight_decay: 1.0e-08
  is_resume: false
  resume_epoch: null
  resume_step: null
  enable_gradient_checkpointing: true
  enable_mixed_precision_training: true
is_resume: false
output_dir: ./results/Checkpoints/0906_internvla_m1