accelerate: compute_environment: LOCAL_MACHINE distributed_type: DEEPSPEED num_machines: 1 num_processes: 2 mixed_precision: bf16 deepspeed_config: zero_stage: 2 gradient_accumulation_steps: 4 gradient_clipping: 1.0 offload_optimizer_device: none offload_param_device: none zero3_init_flag: false zero3_save_16bit_model: false model: architecture: type: dual_system action_dim: 20 bridge_type: cross_attn bridge_layers: - 3 - 7 - 11 - 15 - 19 - 23 - 26 - 29 video_backbone: name: ti2v_5b model_path: /mnt/data/wangyuran/Wan2.2-TI2V-5B action_backbone: dim: 768 ffn_dim: 3072 num_heads: 12 dataloader: type: robotwin dataset_dir: /mnt/data/wangyuran/RoboTwin2.0/dataset task_name: null robot: aloha-agilex variant: both action_mode: eef num_frames: 33 video_stride: 4 height: 480 width: 640 split: train val_ratio: 0.0 repeat: 1 window_stride: 1 multiview: true target_camera: head_camera camera_layout: - head_camera - front_camera - left_camera - right_camera action_stats_path: null max_val_samples: 500 train_tasks: null holdout_tasks: null val_variant: null training_strategy: lambda_video: 1.0 lambda_action: 1.0 freeze: - text_encoder - vae training: debug: false learning_rate: 0.0001 adam_betas: - 0.9 - 0.95 weight_decay: 0.01 max_grad_norm: 1.0 num_epochs: 5 max_steps: null batch_size: 16 gradient_accumulation_steps: 1 lr_scheduler: cosine warmup_ratio: 0.05 lr_min_ratio: 0.01 action_lr: null video_lr: null lora_lr: null use_gradient_checkpointing: true use_gradient_checkpointing_offload: false initialize_model_on_cpu: false max_timestep_boundary: 1.0 min_timestep_boundary: 0.0 lora_base_model: null lora_target_modules: q,k,v,o,ffn.0,ffn.2 lora_rank: 32 output_path: /mnt/data/wangyuran/openwam_checkpoints save_steps: 2000 keep_last_k_ckpts: 3 dataset_num_workers: 4 decoupled: enabled: false video_beta_a: 0.5 video_beta_b: 1.0 warmup_steps: 0 project: name: open-wam output_dir: outputs/${now:%Y-%m-%d_%H-%M-%S} seed: 42 wandb: project: open-wam run_name: null entity: null