| data_root_dir: /home/ubuntu/jianwen-us-midwest-1/tulab/ruisen/.new_maniskill_data | |
| depth: false | |
| global_pose: false | |
| hf_token: .hf_token | |
| image_aug: false | |
| is_grasped: false | |
| is_resume: false | |
| model_type: my_vla_qwen | |
| pretrained_checkpoint: Stanford-ILIAD/prism-qwen25-extra-dinosiglip-224px-0_5b | |
| qpos: false | |
| resume_epoch: null | |
| resume_step: null | |
| run_id: bl_multiview_depth_set_table | |
| run_id_note: null | |
| run_root_dir: myvla_exp | |
| save_interval: 1000 | |
| seed: 7 | |
| segmentation: false | |
| trackers: | |
| - jsonl | |
| - wandb | |
| vla: | |
| action_chunk_size: 8 | |
| action_tokenizer: extra_action_tokenizer | |
| base_vlm: prism-qwen25-extra-dinosiglip-224px+0_5b | |
| compress_history: false | |
| data_mix: bridge | |
| enable_gradient_checkpointing: true | |
| enable_mixed_precision_training: true | |
| epochs: 10 | |
| expected_world_size: 8 | |
| freeze_llm_backbone: false | |
| freeze_vision_backbone: true | |
| global_batch_size: 512 | |
| image_sequence_len: 4 | |
| image_window_size: 1 | |
| learning_rate: 2.0e-05 | |
| lr_scheduler_type: constant | |
| max_grad_norm: 1.0 | |
| max_steps: null | |
| per_device_batch_size: 8 | |
| reduce_in_full_precision: true | |
| save_every_n_steps: 25000 | |
| shuffle_buffer_size: 256000 | |
| train_strategy: fsdp-full-shard | |
| type: myvla-qwen-224px+mx-mshab | |
| unfreeze_last_llm_layer: false | |
| use_flow_matching: false | |
| use_wrist_image: true | |
| vla_id: myvla-qwen-224px+mx-mshab | |
| warmup_ratio: 0.0 | |
| weight_decay: 0.0 | |
| wandb_entity: traysen879-uc-san-diego | |
| wandb_project: mshab_vla | |