| { | |
| "action_dim": 7, | |
| "action_model_type": "DiT-L", | |
| "consolidate_type": "tome", | |
| "dataset_name": "realpushmultit", | |
| "episode_instructions_file": null, | |
| "fusion_type": "gate", | |
| "future_action_window_size": 15, | |
| "group_size": 16, | |
| "hf_token": ".hf_token", | |
| "image_aug": false, | |
| "image_key": "img_third", | |
| "instruction": "push the T-shaped block to visit each T-shaped target region on the table without revisiting any", | |
| "is_resume": true, | |
| "lora": { | |
| "alpha": 48.0, | |
| "cog_cross_targets": [ | |
| "q_proj", | |
| "k_proj", | |
| "v_proj" | |
| ], | |
| "dit_attn_targets": [ | |
| "q", | |
| "v" | |
| ], | |
| "dropout": 0.05, | |
| "enabled": true, | |
| "llama_alpha": 16.0, | |
| "llama_r": 8, | |
| "llama_targets": [ | |
| "q_proj", | |
| "v_proj" | |
| ], | |
| "lora_cog_gate": true, | |
| "lora_llama": true, | |
| "lora_vision": true, | |
| "r": 24, | |
| "vision_alpha": 16.0, | |
| "vision_r": 8, | |
| "vision_targets": [ | |
| "qkv" | |
| ] | |
| }, | |
| "mem_length": 16, | |
| "per_token_size": 256, | |
| "pretrained_checkpoint": "/workspace/diffusion_policy/runs/memoryvla_realpushmultit/memoryvla_realpushmultit_lora_bs64_v1/checkpoints/step-006000-epoch-03-loss=0.0703.pt", | |
| "repeated_diffusion_steps": 4, | |
| "resume_epoch": 3, | |
| "resume_step": 6000, | |
| "retrieval_layers": 2, | |
| "run_id": "memoryvla_realpushmultit_lora_bs64_v1", | |
| "run_id_note": null, | |
| "run_root_dir": "runs/memoryvla_realpushmultit", | |
| "save_interval": 1000, | |
| "seed": 42, | |
| "trackers": [ | |
| "jsonl", | |
| "wandb" | |
| ], | |
| "update_fused": false, | |
| "use_ema": false, | |
| "use_timestep_pe": true, | |
| "val_ratio": 0.05, | |
| "vla": { | |
| "base_vlm": "siglip-224px+7b", | |
| "data_mix": "bridge", | |
| "enable_gradient_checkpointing": true, | |
| "enable_mixed_precision_training": true, | |
| "epochs": 100, | |
| "expected_world_size": 4, | |
| "freeze_llm_backbone": false, | |
| "freeze_vision_backbone": false, | |
| "global_batch_size": 256, | |
| "learning_rate": 0.0002, | |
| "lr_scheduler_type": "linear-warmup+cosine-decay", | |
| "max_grad_norm": 1.0, | |
| "max_steps": 10000, | |
| "per_device_batch_size": 64, | |
| "reduce_in_full_precision": true, | |
| "shuffle_buffer_size": 256000, | |
| "train_strategy": "fsdp-full-shard", | |
| "type": "siglip-224px+mx-bridge", | |
| "unfreeze_last_llm_layer": false, | |
| "vla_id": "siglip-224px+mx-bridge", | |
| "warmup_ratio": 0.05, | |
| "weight_decay": 0.0 | |
| }, | |
| "wandb_entity": "williamcao-uc-san-diego", | |
| "wandb_project": "memoryvla_realpushmultit_lora", | |
| "zarr_path": "data/real_push_multit/RealPushMultiT_320.zarr" | |
| } |