config.json
| 1 | { |
| 2 | "action_horizon": 40, |
| 3 | "add_pos_embed": true, |
| 4 | "apply_sincos_state_encoding": false, |
| 5 | "architectures": [ |
| 6 | "Gr00tN1d7" |
| 7 | ], |
| 8 | "attn_dropout": 0.2, |
| 9 | "backbone_embedding_dim": 2048, |
| 10 | "color_jitter_params": { |
| 11 | "brightness": 0.3, |
| 12 | "contrast": 0.4, |
| 13 | "hue": 0.08, |
| 14 | "saturation": 0.5 |
| 15 | }, |
| 16 | "crop_fraction": 0.95, |
| 17 | "diffusion_model_cfg": { |
| 18 | "attention_head_dim": 48, |
| 19 | "dropout": 0.2, |
| 20 | "final_dropout": true, |
| 21 | "interleave_self_attention": true, |
| 22 | "norm_type": "ada_norm", |
| 23 | "num_attention_heads": 32, |
| 24 | "num_layers": 32, |
| 25 | "output_dim": 1024, |
| 26 | "positional_embeddings": null |
| 27 | }, |
| 28 | "dtype": "bfloat16", |
| 29 | "exclude_state": false, |
| 30 | "formalize_language": true, |
| 31 | "hidden_size": 1024, |
| 32 | "image_crop_size": [ |
| 33 | 230, |
| 34 | 230 |
| 35 | ], |
| 36 | "image_target_size": [ |
| 37 | 256, |
| 38 | 256 |
| 39 | ], |
| 40 | "letter_box_transform": false, |
| 41 | "load_bf16": true, |
| 42 | "max_action_dim": 132, |
| 43 | "max_num_embodiments": 32, |
| 44 | "max_seq_len": 1024, |
| 45 | "max_state_dim": 132, |
| 46 | "model_dtype": "bfloat16", |
| 47 | "model_type": "Gr00tN1d7", |
| 48 | "noise_beta_alpha": 1.5, |
| 49 | "noise_beta_beta": 1.0, |
| 50 | "noise_s": 0.999, |
| 51 | "num_inference_timesteps": 4, |
| 52 | "num_timestep_buckets": 1000, |
| 53 | "random_history_crop": true, |
| 54 | "random_rotation_angle": 0, |
| 55 | "reproject_vision": false, |
| 56 | "rtc_ramp_rate": 6.0, |
| 57 | "select_layer": 16, |
| 58 | "shortest_image_edge": 256, |
| 59 | "state_dropout_prob": 0.2, |
| 60 | "state_gaussian_noise_std": 0.0, |
| 61 | "transformers_version": "4.57.1", |
| 62 | "tune_diffusion_model": true, |
| 63 | "tune_linear": true, |
| 64 | "tune_llm": true, |
| 65 | "tune_projector": true, |
| 66 | "tune_top_llm_layers": 0, |
| 67 | "tune_visual": true, |
| 68 | "tune_vlln": true, |
| 69 | "use_albumentations": true, |
| 70 | "use_alternate_vl_dit": true, |
| 71 | "use_flash_attention": true, |
| 72 | "use_future_tokens": false, |
| 73 | "use_mean_std": false, |
| 74 | "use_percentiles": true, |
| 75 | "use_vl_self_attention": true, |
| 76 | "use_vlln": true, |
| 77 | "vl_self_attention_cfg": { |
| 78 | "attention_head_dim": 64, |
| 79 | "dropout": 0.2, |
| 80 | "final_dropout": true, |
| 81 | "num_attention_heads": 32, |
| 82 | "num_layers": 4, |
| 83 | "positional_embeddings": null |
| 84 | }, |
| 85 | "model_name": "nvidia/Cosmos-Reason2-2B" |
| 86 | } |
| 87 | |