Upload config.json with huggingface_hub
Browse files- config.json +137 -0
config.json
ADDED
|
@@ -0,0 +1,137 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"action_horizon": 16,
|
| 3 |
+
"action_model_lora_alpha": 32,
|
| 4 |
+
"action_model_lora_dropout": 0.0,
|
| 5 |
+
"action_model_lora_rank": 16,
|
| 6 |
+
"action_model_lora_target_modules": [
|
| 7 |
+
"vl_qkv",
|
| 8 |
+
"vl_proj",
|
| 9 |
+
"sa_qkv",
|
| 10 |
+
"sa_proj",
|
| 11 |
+
"p_qkv",
|
| 12 |
+
"p_proj",
|
| 13 |
+
"linear1",
|
| 14 |
+
"linear2"
|
| 15 |
+
],
|
| 16 |
+
"action_model_use_lora": false,
|
| 17 |
+
"actlat_embodiment_id": "openarm_prq",
|
| 18 |
+
"actlat_mode": false,
|
| 19 |
+
"actlat_prq_stats_path": "",
|
| 20 |
+
"actlat_standardize": false,
|
| 21 |
+
"actlat_tokenizer_path": "",
|
| 22 |
+
"actlat_z_stats_path": "",
|
| 23 |
+
"anchor_dim_norm": "",
|
| 24 |
+
"anchor_dim_stats_path": "",
|
| 25 |
+
"anchor_human_tips_weight": 1.0,
|
| 26 |
+
"anchor_readout_asym": false,
|
| 27 |
+
"anchor_readout_dim": 24,
|
| 28 |
+
"anchor_readout_shuffle": false,
|
| 29 |
+
"anchor_readout_taps": "cog,d2,s3,s7",
|
| 30 |
+
"anchor_readout_tweight": true,
|
| 31 |
+
"anchor_readout_weight": 0.0,
|
| 32 |
+
"anchor_tips_dims": 15,
|
| 33 |
+
"architectures": [
|
| 34 |
+
"RLDX"
|
| 35 |
+
],
|
| 36 |
+
"backbone_lora_alpha": 32,
|
| 37 |
+
"backbone_lora_dropout": 0.0,
|
| 38 |
+
"backbone_lora_num_layers": -1,
|
| 39 |
+
"backbone_lora_rank": 16,
|
| 40 |
+
"backbone_lora_target_modules": [
|
| 41 |
+
"q_proj",
|
| 42 |
+
"k_proj",
|
| 43 |
+
"v_proj",
|
| 44 |
+
"o_proj",
|
| 45 |
+
"gate_proj",
|
| 46 |
+
"up_proj",
|
| 47 |
+
"down_proj"
|
| 48 |
+
],
|
| 49 |
+
"backbone_use_lora": false,
|
| 50 |
+
"color_jitter_params": null,
|
| 51 |
+
"conversation_image_first": false,
|
| 52 |
+
"diffusion_model_cfg": {
|
| 53 |
+
"action_model_max_seq_len": 512,
|
| 54 |
+
"attention_head_dim": 64,
|
| 55 |
+
"depth_multi_stream": 4,
|
| 56 |
+
"depth_single_stream": 8,
|
| 57 |
+
"dropout": 0.2,
|
| 58 |
+
"final_dropout": true,
|
| 59 |
+
"num_attention_heads": 24,
|
| 60 |
+
"output_dim": 1024,
|
| 61 |
+
"physics_dim": 0,
|
| 62 |
+
"positional_embeddings": "rope_sa_only",
|
| 63 |
+
"pre_norm": "layer_norm",
|
| 64 |
+
"qk_norm": "rms_norm",
|
| 65 |
+
"rope_theta": 10000.0,
|
| 66 |
+
"sa_dim": 1536,
|
| 67 |
+
"temb_type": "input_token",
|
| 68 |
+
"use_physics": false,
|
| 69 |
+
"use_swiglu": true,
|
| 70 |
+
"vl_dim": 4096
|
| 71 |
+
},
|
| 72 |
+
"dtype": "bfloat16",
|
| 73 |
+
"egopi_split_embodiment": false,
|
| 74 |
+
"embod_reg_gather": false,
|
| 75 |
+
"embod_reg_hand_bins": 0,
|
| 76 |
+
"embod_reg_hand_edges": "",
|
| 77 |
+
"embod_reg_layer": 2,
|
| 78 |
+
"embod_reg_mode": "",
|
| 79 |
+
"embod_reg_tbins": 0,
|
| 80 |
+
"embod_reg_vic_cov": 0.04,
|
| 81 |
+
"embod_reg_vic_var": 1.0,
|
| 82 |
+
"embod_reg_weight": 0.05,
|
| 83 |
+
"freeze_cog_tokens": false,
|
| 84 |
+
"freeze_modules": "",
|
| 85 |
+
"general_embodiment_train_ratio": 0,
|
| 86 |
+
"gr1hp_align_dim": 24,
|
| 87 |
+
"gr1hp_align_only": true,
|
| 88 |
+
"gr1hp_align_shuffle": false,
|
| 89 |
+
"gr1hp_align_weight": 0.2,
|
| 90 |
+
"gr1hp_split_embodiment": false,
|
| 91 |
+
"image_max_area": 65536,
|
| 92 |
+
"image_resize_m": 32,
|
| 93 |
+
"keep_full_lm": false,
|
| 94 |
+
"memory_cfg": {
|
| 95 |
+
"hidden_size": 4096,
|
| 96 |
+
"intermediate_size": 16384,
|
| 97 |
+
"max_position_embeddings": 32,
|
| 98 |
+
"num_attention_heads": 16,
|
| 99 |
+
"num_hidden_layers": 2,
|
| 100 |
+
"num_key_value_heads": 16,
|
| 101 |
+
"rms_norm_eps": 1e-05,
|
| 102 |
+
"use_causal_attn": true,
|
| 103 |
+
"use_rope": true
|
| 104 |
+
},
|
| 105 |
+
"model_name": "RLWRLD/RLDX-1-VLM",
|
| 106 |
+
"model_revision": null,
|
| 107 |
+
"model_type": "RLDX-1",
|
| 108 |
+
"n_cog_tokens": 64,
|
| 109 |
+
"physics_delta_indices": null,
|
| 110 |
+
"physics_dims": [],
|
| 111 |
+
"physics_keys": [],
|
| 112 |
+
"physics_use_flow_matching": true,
|
| 113 |
+
"random_crop_fraction": null,
|
| 114 |
+
"random_rotation_angle": null,
|
| 115 |
+
"rtc_inference_delay": 0,
|
| 116 |
+
"rtc_inference_exec_horizon": 0,
|
| 117 |
+
"rtc_inference_mode": "none",
|
| 118 |
+
"rtc_jacobian_steps_only": 3,
|
| 119 |
+
"rtc_training_max_delay": 0,
|
| 120 |
+
"stage_from_checkpoint": "",
|
| 121 |
+
"state_dropout_prob": 0.0,
|
| 122 |
+
"train_stage": "",
|
| 123 |
+
"transformers_version": "4.57.0",
|
| 124 |
+
"tune_diffusion_model": true,
|
| 125 |
+
"tune_llm": false,
|
| 126 |
+
"tune_projector": true,
|
| 127 |
+
"tune_top_llm_layers": 0,
|
| 128 |
+
"tune_visual": false,
|
| 129 |
+
"use_memory": false,
|
| 130 |
+
"use_motion": false,
|
| 131 |
+
"use_physics": false,
|
| 132 |
+
"use_video": true,
|
| 133 |
+
"video_length": 1,
|
| 134 |
+
"video_stride": 2,
|
| 135 |
+
"vlm_lang_loss": false,
|
| 136 |
+
"vlm_lang_weight": 0.5
|
| 137 |
+
}
|