Upload ftl_cube_v8/experiment_cfg/final_model_config.json with huggingface_hub
Browse files
ftl_cube_v8/experiment_cfg/final_model_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_type": "Gr00tN1d7",
|
| 3 |
+
"model_dtype": "bfloat16",
|
| 4 |
+
"model_name": "nvidia/Cosmos-Reason2-2B",
|
| 5 |
+
"backbone_model_type": "qwen",
|
| 6 |
+
"model_revision": null,
|
| 7 |
+
"tune_top_llm_layers": 0,
|
| 8 |
+
"backbone_embedding_dim": 2048,
|
| 9 |
+
"tune_llm": false,
|
| 10 |
+
"tune_visual": false,
|
| 11 |
+
"select_layer": 16,
|
| 12 |
+
"reproject_vision": false,
|
| 13 |
+
"use_flash_attention": true,
|
| 14 |
+
"load_bf16": false,
|
| 15 |
+
"backbone_trainable_params_fp32": true,
|
| 16 |
+
"letter_box_transform": false,
|
| 17 |
+
"extra_augmentation_config": null,
|
| 18 |
+
"apply_sincos_state_encoding": false,
|
| 19 |
+
"use_percentiles": true,
|
| 20 |
+
"use_relative_action": false,
|
| 21 |
+
"max_state_dim": 132,
|
| 22 |
+
"max_action_dim": 132,
|
| 23 |
+
"action_horizon": 40,
|
| 24 |
+
"hidden_size": 1024,
|
| 25 |
+
"input_embedding_dim": 1536,
|
| 26 |
+
"state_history_length": 1,
|
| 27 |
+
"add_pos_embed": true,
|
| 28 |
+
"attn_dropout": 0.2,
|
| 29 |
+
"use_vlln": true,
|
| 30 |
+
"max_seq_len": 1024,
|
| 31 |
+
"use_alternate_vl_dit": true,
|
| 32 |
+
"attend_text_every_n_blocks": 2,
|
| 33 |
+
"diffusion_model_cfg": {
|
| 34 |
+
"attention_head_dim": 48,
|
| 35 |
+
"dropout": 0.2,
|
| 36 |
+
"final_dropout": true,
|
| 37 |
+
"interleave_self_attention": true,
|
| 38 |
+
"norm_type": "ada_norm",
|
| 39 |
+
"num_attention_heads": 32,
|
| 40 |
+
"num_layers": 32,
|
| 41 |
+
"output_dim": 1024,
|
| 42 |
+
"positional_embeddings": null
|
| 43 |
+
},
|
| 44 |
+
"ttt_traj_len": null,
|
| 45 |
+
"ttt_demo_steps": 0,
|
| 46 |
+
"num_inference_timesteps": 4,
|
| 47 |
+
"noise_beta_alpha": 1.5,
|
| 48 |
+
"noise_beta_beta": 1.0,
|
| 49 |
+
"noise_s": 0.999,
|
| 50 |
+
"num_timestep_buckets": 1000,
|
| 51 |
+
"tune_projector": true,
|
| 52 |
+
"tune_diffusion_model": true,
|
| 53 |
+
"tune_vlln": true,
|
| 54 |
+
"state_dropout_prob": 0.0,
|
| 55 |
+
"exclude_state": false,
|
| 56 |
+
"use_mean_std": false,
|
| 57 |
+
"max_num_embodiments": 32
|
| 58 |
+
}
|