| { |
| "step": 80000, |
| "args": { |
| "base_model": "/data/base_model", |
| "config_model": "/data/config_model", |
| "dataset": "/data/dataset", |
| "vl_model_path": null, |
| "canonical_vl_model_name": "nvidia/Cosmos-Reason2-2B", |
| "label_dir": "adaptive_speed_labels", |
| "output_dir": "/data/output", |
| "experiment_name": "task0003_vlmmerged_adaptive_speed_chunk32_a2_m1_80k_gpu0_20260811_092736", |
| "max_steps": 80000, |
| "global_batch_size": 8, |
| "gradient_accumulation_steps": 1, |
| "num_workers": 4, |
| "actor_lr": 0.0001, |
| "weight_decay": 1e-05, |
| "warmup_ratio": 0.05, |
| "action_horizon": 32, |
| "approach_speed_min": 2.0, |
| "approach_speed_max": 2.0, |
| "transition_speed": 1.0, |
| "manipulation_speed": 1.0, |
| "approach_token_weight": 1.0, |
| "transition_token_weight": 1.15, |
| "manipulation_token_weight": 1.4, |
| "approach_gripper_weight": 0.6, |
| "transition_gripper_weight": 1.5, |
| "manipulation_gripper_weight": 3.0, |
| "arm_weight": 1.0, |
| "gripper_indices": "7,15", |
| "approach_phase_id": 0, |
| "transition_phase_id": 1, |
| "manipulation_phase_id": 2, |
| "episode_sampling_rate": 0.1, |
| "shard_size": 1024, |
| "num_shards_per_epoch": 100000, |
| "save_steps": 20000, |
| "save_at_steps": "60000,80000", |
| "logging_steps": 10, |
| "max_grad_norm": 1.0, |
| "seed": 42, |
| "bf16": true, |
| "tf32": true, |
| "local_files_only": true, |
| "use_percentiles": false, |
| "clip_outliers": true, |
| "state_dropout_prob": 0.2, |
| "rtc_training_prefix_steps": 0, |
| "save_optimizer_state": false, |
| "save_final_model": false, |
| "allow_terminal_padding": true, |
| "dry_run_data": false, |
| "dry_run_forward": false |
| } |
| } |
