Instructions to use zaleni/WSA-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use zaleni/WSA-Base with LeRobot:
- Notebooks
- Google Colab
- Kaggle
| { | |
| "type": "TBot_SA1", | |
| "n_obs_steps": 1, | |
| "input_features": { | |
| "observation.state": { | |
| "type": "STATE", | |
| "shape": [ | |
| 32 | |
| ] | |
| } | |
| }, | |
| "output_features": { | |
| "action": { | |
| "type": "ACTION", | |
| "shape": [ | |
| 32 | |
| ] | |
| } | |
| }, | |
| "device": "cuda:0", | |
| "use_amp": false, | |
| "push_to_hub": false, | |
| "repo_id": "zaleni/TBot-SA1-Base", | |
| "private": null, | |
| "tags": null, | |
| "license": null, | |
| "pretrained_path": null, | |
| "qwen3_vl_variant": "qwen3_vl_28l", | |
| "action_expert_variant": "qwen3_28l", | |
| "qwen3_vl_pretrained_path": "Qwen/Qwen3-VL-2B-Instruct", | |
| "dtype": "bfloat16", | |
| "chunk_size": 50, | |
| "n_action_steps": 50, | |
| "max_state_dim": 32, | |
| "max_action_dim": 32, | |
| "num_inference_steps": 10, | |
| "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, | |
| "image_resolution": [ | |
| 224, | |
| 224 | |
| ], | |
| "empty_cameras": 0, | |
| "normalization_mapping": { | |
| "VISUAL": "IDENTITY", | |
| "STATE": "IDENTITY", | |
| "ACTION": "IDENTITY" | |
| }, | |
| "gradient_checkpointing": false, | |
| "compile_model": false, | |
| "compile_mode": "max-autotune", | |
| "optimizer_lr": 5e-05, | |
| "optimizer_betas": [ | |
| 0.9, | |
| 0.95 | |
| ], | |
| "optimizer_eps": 1e-08, | |
| "optimizer_weight_decay": 0.01, | |
| "optimizer_grad_clip_norm": 1.0, | |
| "scheduler_warmup_steps": 2000, | |
| "scheduler_decay_steps": 300000, | |
| "scheduler_decay_lr": 1e-05, | |
| "tokenizer_max_length": 48, | |
| "freeze_vision_encoder": false, | |
| "train_expert_only": false, | |
| "train_vlm_only": false, | |
| "scale_factor": 8, | |
| "lambda_gen": 0.01, | |
| "cosmos_tokenizer_path_or_name": "nvidia/Cosmos-Tokenizer-CI8x8", | |
| "enable_3d_queries": true, | |
| "num_3d_query_tokens": 432, | |
| "da3_alignment_mode": "query_decoder", | |
| "da3_query_resampler_layers": 1, | |
| "da3_query_resampler_ff_mult": 1, | |
| "query_layer_indices": [ | |
| 13, | |
| 19, | |
| 23, | |
| 27 | |
| ], | |
| "da3_variant": "auto", | |
| "da3_teacher_layers": [ | |
| 11, | |
| 15, | |
| 19, | |
| 23 | |
| ], | |
| "da3_query_dim": 2048, | |
| "da3_tokens_per_view": 1296, | |
| "da3_num_views": 3, | |
| "lambda_3d": 0.01, | |
| "da3_model_path_or_name": "depth-anything/DA3-LARGE-1.1", | |
| "da3_model_name": null, | |
| "da3_code_root": null, | |
| "da3_teacher_process_res": 504, | |
| "da3_layer_weights": [ | |
| 1.0, | |
| 1.2, | |
| 1.4, | |
| 1.6 | |
| ], | |
| "future_query_init_std": 0.02, | |
| "log_da3_teacher_timing": true | |
| } |