| run_id: starvla_qwen3oft_fourier_gr1_unified_1000_franka_epfrac_1_3 |
| run_root_dir: ./playground/Checkpoints |
| seed: 42 |
| wandb_entity: zwanggk |
| wandb_project: sq_robocasa |
| is_debug: false |
| version_id: '0.21' |
| framework: |
| name: QwenOFT |
| qwenvl: |
| base_vlm: /gpfs/wangzixuan/shared_project/sq/5-1/starVLA/playground/Pretrained_models/Qwen3-VL-4B-Instruct-Action |
| attn_implementation: flash_attention_2 |
| vl_hidden_dim: 2048 |
| action_model: |
| action_model_type: DiT-B |
| action_dim: 29 |
| action_hidden_dim: 2560 |
| future_action_window_size: 15 |
| past_action_window_size: 0 |
| hidden_size: 2560 |
| add_pos_embed: true |
| max_seq_len: 1024 |
| state_dim: 58 |
| action_horizon: 16 |
| repeated_diffusion_steps: 8 |
| noise_beta_alpha: 1.5 |
| noise_beta_beta: 1.0 |
| noise_s: 0.999 |
| num_timestep_buckets: 1000 |
| num_inference_timesteps: 4 |
| num_target_vision_tokens: 32 |
| diffusion_model_cfg: |
| dropout: 0.2 |
| final_dropout: true |
| interleave_self_attention: true |
| norm_type: ada_norm |
| num_layers: 16 |
| positional_embeddings: null |
| output_dim: 2560 |
| cross_attention_dim: 2048 |
| dino: |
| dino_backbone: dinov2_vits14 |
| datasets: |
| vlm_data: |
| dataset_py: vlm_datasets |
| dataformat: llava_json |
| dataset_use: asv2_conversation_en,asv2_detailed_description_en,asv2_region_captioning_en,coco_internvl_longcap_en,coco_karpathy_train_567_en,coco_negative_gpt4o_en,coco_poetry_zh,coco_rem_en_zh,cocorem_exist_yorn_en,cocotextv2_en,cocotextv2_gpt4o_en,okvqa_en,refcoco_grounding_aug_en,refcoco_grounding_en,tallyqa_coco_en,toloka_grounding_aug_en,vqav2_en,vsr_en |
| eval_dataset: aokvqa_cauldron_llava_format |
| data_flatten: false |
| base_interval: 2 |
| max_pixels: 50176 |
| min_pixels: 784 |
| model_max_length: 2048 |
| model_type: qwen2.5vl |
| per_device_batch_size: 4 |
| vla_data: |
| dataset_py: lerobot_datasets |
| include_state: false |
| data_root_dir: /gpfs/wangzixuan/visual_prompting/starVLA_robocasa/playground/Datasets/nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim |
| data_mix: fourier_gr1_unified_1000 |
| action_type: delta_ee |
| sequential_step_sampling: false |
| CoT_prompt: Your task is {instruction}. To identify the key objects for your task. |
| Locate their bounding boxes in [x1,y1,x2,y2] format. |
| per_device_batch_size: 32 |
| load_all_data_for_training: true |
| obs_image_size: |
| - 224 |
| - 224 |
| delete_pause_frame: false |
| episode_subset_fraction: 0.3333333333 |
| episode_subset_seed: 42 |
| trainer: |
| max_train_steps: 100000 |
| num_warmup_steps: 5000 |
| save_interval: 10000 |
| eval_interval: 100 |
| learning_rate: |
| base: 3.0e-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: true |
| loss_scale: |
| vla: 1.0 |
| vlm: 0.1 |
| max_grad_norm: 1.0 |
| 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 |
| reload_modules: qwen_vl_interface |
| pretrained_checkpoint: /gpfs/wangzixuan/shared_project/sq/5-1/starVLA/playground/Pretrained_models/0426_droid_pretrain_qwen3GR00T_triple_oft_pi_50k_16BS/final_model/pytorch_model.pt |
| config_yaml: ./examples/Robocasa_tabletop/train_files/starvla_cotrain_robocasa_gr1.yaml |
| output_dir: ./playground/Checkpoints/starvla_qwen3oft_fourier_gr1_unified_1000_franka_epfrac_1_3 |
|
|
