| { |
| "type": "pi05_rl", |
| "n_obs_steps": 1, |
| "input_features": { |
| "observation.images.side": { |
| "type": "VISUAL", |
| "shape": [ |
| 3, |
| 224, |
| 224 |
| ] |
| }, |
| "observation.images.top": { |
| "type": "VISUAL", |
| "shape": [ |
| 3, |
| 224, |
| 224 |
| ] |
| }, |
| "observation.state": { |
| "type": "STATE", |
| "shape": [ |
| 6 |
| ] |
| } |
| }, |
| "output_features": { |
| "action": { |
| "type": "ACTION", |
| "shape": [ |
| 6 |
| ] |
| } |
| }, |
| "device": "cuda", |
| "use_amp": false, |
| "use_peft": false, |
| "push_to_hub": true, |
| "repo_id": "cijerezg/multi-task-toys-merged-v2", |
| "private": null, |
| "tags": null, |
| "license": null, |
| "pretrained_path": null, |
| "paligemma_variant": "gemma_2b", |
| "action_expert_variant": "gemma_300m", |
| "dtype": "bfloat16", |
| "chunk_size": 50, |
| "n_action_steps": 50, |
| "max_state_dim": 6, |
| "max_action_dim": 32, |
| "num_inference_steps": 5, |
| "time_sampling_beta_alpha": 1.5, |
| "time_sampling_beta_beta": 1.0, |
| "time_sampling_scale": 0.999, |
| "time_sampling_offset": 0.001, |
| "min_period": 0.004, |
| "max_period": 4.0, |
| "rtc_config": { |
| "enabled": true, |
| "prefix_attention_schedule": "LINEAR", |
| "max_guidance_weight": 10.0, |
| "execution_horizon": 10, |
| "debug": false, |
| "debug_maxlen": 100 |
| }, |
| "image_resolution": [ |
| 224, |
| 224 |
| ], |
| "empty_cameras": 0, |
| "use_dataset_stats": false, |
| "normalization_mapping": { |
| "VISUAL": "IDENTITY", |
| "STATE": "MIN_MAX", |
| "ENV": "MIN_MAX", |
| "ACTION": "QUANTILES" |
| }, |
| "action_tokenizer_name": "physical-intelligence/fast", |
| "text_tokenizer_name": "google/paligemma-3b-pt-224", |
| "max_action_tokens": 256, |
| "fast_skip_tokens": 128, |
| "max_decoding_steps": 200, |
| "temperature": 0.0, |
| "subtask_regeneration_interval": 1.5, |
| "gradient_checkpointing": true, |
| "compile_model": false, |
| "compile_mode": "max-autotune", |
| "freeze_vision_encoder": false, |
| "train_expert_only": false, |
| "knowledge_insulation": true, |
| "action_encoding": "anchor", |
| "loss_weight_flow": 1.0, |
| "loss_weight_action_ce": 1.0, |
| "loss_weight_subtask_ce": 1.0, |
| "optimizer_lr": 2.5e-05, |
| "optimizer_betas": [ |
| 0.9, |
| 0.95 |
| ], |
| "optimizer_eps": 1e-08, |
| "optimizer_weight_decay": 0.1, |
| "optimizer_grad_clip_norm": 1.0, |
| "scheduler_warmup_steps": 1000, |
| "scheduler_decay_steps": 30000, |
| "scheduler_decay_lr": 2.5e-06, |
| "tokenizer_max_length": 64, |
| "task": "Pick up the orange cube and place it on the black X marker", |
| "action_dim": 6, |
| "drop_n_last_frames": 2, |
| "critic_target_update_weight": 0.005, |
| "num_critics": 1, |
| "discount": 0.97, |
| "reward_normalization_constant": 5.0, |
| "terminal_failure_reward": -16.0, |
| "online_steps": 20000, |
| "online_buffer_capacity": 5000, |
| "offline_buffer_capacity": 50000, |
| "async_prefetch": false, |
| "online_step_before_learning": 10, |
| "policy_update_freq": 1, |
| "grad_clip_norm": 2.0, |
| "gradient_accumulation_steps": 16, |
| "critic_lr": 5e-05, |
| "actor_lr": 5e-05, |
| "utd_ratio": 1, |
| "actor_device": "cuda:0", |
| "learner_device": "cuda:0", |
| "use_separate_critic": true, |
| "critic_llm_depth": 6, |
| "critic_network_kwargs": { |
| "hidden_dims": [ |
| 256, |
| 256 |
| ], |
| "activate_final": true |
| }, |
| "trainable_params": { |
| "vision_encoder_from_layer": { |
| "vision_tower": 5, |
| "multi_modal_projector": true |
| }, |
| "language_from_layer": 0, |
| "critic_language_from_layer": 5 |
| }, |
| "offline_steps": 10000, |
| "inference_advantage": 1.0, |
| "advantage_scaling": 0.2, |
| "pi05_checkpoint": "outputs/pi05_base", |
| "action_encoding_stats_path": "outputs/stats_jack/action_stats_anchor_jack_cube.pt", |
| "dataset_stats": null, |
| "storage_device": "cpu", |
| "shared_encoder": false, |
| "num_discrete_actions": null, |
| "vision_encoder_name": null, |
| "actor_learner_config": { |
| "learner_host": "192.168.50.1", |
| "learner_port": 50051, |
| "policy_parameters_push_frequency": 180, |
| "queue_get_timeout": 2 |
| }, |
| "concurrency": { |
| "actor": "threads", |
| "learner": "threads" |
| } |
| } |
