Upload config.json
Browse files- config.json +42 -0
config.json
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_gradient_checkpointing": true,
|
| 3 |
+
"_name_or_path": "Yysrc/LIBERO-Object",
|
| 4 |
+
"action_dim": 7,
|
| 5 |
+
"action_model_type": "DiT-L",
|
| 6 |
+
"architectures": [
|
| 7 |
+
"Mantis"
|
| 8 |
+
],
|
| 9 |
+
"attn_implementation": null,
|
| 10 |
+
"connector_num_hidden_layers": 12,
|
| 11 |
+
"diffusion_model_id": "Efficient-Large-Model/Sana_600M_512px_diffusers",
|
| 12 |
+
"future_action_window_size": 4,
|
| 13 |
+
"in_channels": 32,
|
| 14 |
+
"input_size": 16,
|
| 15 |
+
"loss_type": "flow",
|
| 16 |
+
"max_input_text_tokens": 256,
|
| 17 |
+
"max_timestep_gap": 6,
|
| 18 |
+
"mllm_id": "Qwen/Qwen2.5-VL-3B-Instruct",
|
| 19 |
+
"model_type": "mantis",
|
| 20 |
+
"modules_to_freeze": [
|
| 21 |
+
"vae"
|
| 22 |
+
],
|
| 23 |
+
"modules_to_unfreeze": [
|
| 24 |
+
"model.mllm_backbone.model.embed_tokens",
|
| 25 |
+
"model.policy_head",
|
| 26 |
+
"model.transformer",
|
| 27 |
+
"model.connector",
|
| 28 |
+
"model.mllm_backbone"
|
| 29 |
+
],
|
| 30 |
+
"noise_scheduler_id": "Efficient-Large-Model/Sana_600M_512px_diffusers",
|
| 31 |
+
"num_actqueries": 6,
|
| 32 |
+
"num_gapqueries": 3,
|
| 33 |
+
"num_metaqueries": 9,
|
| 34 |
+
"past_action_window_size": 0,
|
| 35 |
+
"scheduler_id": "Efficient-Large-Model/Sana_600M_512px_diffusers",
|
| 36 |
+
"system_prompt": "You will be provided with an image observation and a corresponding instruction.",
|
| 37 |
+
"torch_dtype": "float32",
|
| 38 |
+
"training_mode": "image_action",
|
| 39 |
+
"transformers_version": "4.49.0",
|
| 40 |
+
"vae_downsample_f": 32,
|
| 41 |
+
"vae_id": "Efficient-Large-Model/Sana_600M_512px_diffusers"
|
| 42 |
+
}
|