Upload fine-tuned GR00T model for stack cube task
Browse files- README.md +57 -0
- config.json +64 -0
- experiment_cfg/metadata.json +243 -0
- model-00001-of-00002.safetensors +3 -0
- model-00002-of-00002.safetensors +3 -0
- model.safetensors.index.json +0 -0
README.md
ADDED
|
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- robotics
|
| 5 |
+
- embodied-ai
|
| 6 |
+
- imitation-learning
|
| 7 |
+
- groot
|
| 8 |
+
- easymimic
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# EasyMimic Stack Cube Task - Fine-tuned GR00T Model
|
| 12 |
+
|
| 13 |
+
这是基于 GR00T-N1.5-3B 在 "把一个方块堆在另一个方块上" 任务上微调的模型。
|
| 14 |
+
|
| 15 |
+
## 模型信息
|
| 16 |
+
|
| 17 |
+
- **基础模型**: GR00T-N1.5-3B
|
| 18 |
+
- **任务**: Stack one cube on another cube
|
| 19 |
+
- **训练数据**:
|
| 20 |
+
- 人手演示: 110 episodes (11 videos × 10 segments)
|
| 21 |
+
- 机械臂演示: 20 episodes (2 videos × 10 segments)
|
| 22 |
+
- **训练步数**: 3000 steps
|
| 23 |
+
- **最终损失**: 0.0326
|
| 24 |
+
|
| 25 |
+
## 训练配置
|
| 26 |
+
|
| 27 |
+
- Batch size: 8
|
| 28 |
+
- Learning rate: 5e-5
|
| 29 |
+
- 优化器: AdamW
|
| 30 |
+
- 微调组件: Projector + Diffusion Model
|
| 31 |
+
- 数据平衡: 启用 dataset weights 和 trajectory weights 平衡
|
| 32 |
+
|
| 33 |
+
## 使用方法
|
| 34 |
+
|
| 35 |
+
```python
|
| 36 |
+
from transformers import AutoModel, AutoTokenizer
|
| 37 |
+
|
| 38 |
+
model = AutoModel.from_pretrained("paradise-wujie/easymimic-stack-cube-groot")
|
| 39 |
+
tokenizer = AutoTokenizer.from_pretrained("paradise-wujie/easymimic-stack-cube-groot")
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
## 训练日志
|
| 43 |
+
|
| 44 |
+
完整训练日志和代码请查看: https://github.com/KKqdtjo/MyEasyMimic
|
| 45 |
+
|
| 46 |
+
## 引用
|
| 47 |
+
|
| 48 |
+
如果使用此模型,请引用 EasyMimic 论文:
|
| 49 |
+
|
| 50 |
+
```bibtex
|
| 51 |
+
@article{easymimic2024,
|
| 52 |
+
title={EasyMimic: Learning Robotic Manipulation from Human Demonstrations},
|
| 53 |
+
author={...},
|
| 54 |
+
journal={...},
|
| 55 |
+
year={2024}
|
| 56 |
+
}
|
| 57 |
+
```
|
config.json
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"action_dim": 32,
|
| 3 |
+
"action_head_cfg": {
|
| 4 |
+
"action_dim": 32,
|
| 5 |
+
"action_horizon": 16,
|
| 6 |
+
"add_pos_embed": true,
|
| 7 |
+
"backbone_embedding_dim": 2048,
|
| 8 |
+
"diffusion_model_cfg": {
|
| 9 |
+
"attention_head_dim": 48,
|
| 10 |
+
"cross_attention_dim": 2048,
|
| 11 |
+
"dropout": 0.2,
|
| 12 |
+
"final_dropout": true,
|
| 13 |
+
"interleave_self_attention": true,
|
| 14 |
+
"norm_type": "ada_norm",
|
| 15 |
+
"num_attention_heads": 32,
|
| 16 |
+
"num_layers": 16,
|
| 17 |
+
"output_dim": 1024,
|
| 18 |
+
"positional_embeddings": null
|
| 19 |
+
},
|
| 20 |
+
"hidden_size": 1024,
|
| 21 |
+
"input_embedding_dim": 1536,
|
| 22 |
+
"max_action_dim": 32,
|
| 23 |
+
"max_state_dim": 64,
|
| 24 |
+
"model_dtype": "float32",
|
| 25 |
+
"noise_beta_alpha": 1.5,
|
| 26 |
+
"noise_beta_beta": 1.0,
|
| 27 |
+
"noise_s": 0.999,
|
| 28 |
+
"num_inference_timesteps": 4,
|
| 29 |
+
"num_target_vision_tokens": 32,
|
| 30 |
+
"num_timestep_buckets": 1000,
|
| 31 |
+
"tune_diffusion_model": true,
|
| 32 |
+
"tune_projector": true,
|
| 33 |
+
"use_vlln": true,
|
| 34 |
+
"vl_self_attention_cfg": {
|
| 35 |
+
"attention_head_dim": 64,
|
| 36 |
+
"dropout": 0.2,
|
| 37 |
+
"final_dropout": true,
|
| 38 |
+
"num_attention_heads": 32,
|
| 39 |
+
"num_layers": 4,
|
| 40 |
+
"positional_embeddings": null
|
| 41 |
+
}
|
| 42 |
+
},
|
| 43 |
+
"action_horizon": 16,
|
| 44 |
+
"architectures": [
|
| 45 |
+
"GR00T_N1_5"
|
| 46 |
+
],
|
| 47 |
+
"attn_implementation": null,
|
| 48 |
+
"backbone_cfg": {
|
| 49 |
+
"eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
|
| 50 |
+
"load_bf16": false,
|
| 51 |
+
"project_to_dim": null,
|
| 52 |
+
"reproject_vision": false,
|
| 53 |
+
"select_layer": 12,
|
| 54 |
+
"tune_llm": false,
|
| 55 |
+
"tune_visual": true,
|
| 56 |
+
"use_flash_attention": true
|
| 57 |
+
},
|
| 58 |
+
"compute_dtype": "bfloat16",
|
| 59 |
+
"hidden_size": 2048,
|
| 60 |
+
"model_dtype": "float32",
|
| 61 |
+
"model_type": "gr00t_n1_5",
|
| 62 |
+
"torch_dtype": "bfloat16",
|
| 63 |
+
"transformers_version": "4.51.3"
|
| 64 |
+
}
|
experiment_cfg/metadata.json
ADDED
|
@@ -0,0 +1,243 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"new_embodiment": {
|
| 3 |
+
"statistics": {
|
| 4 |
+
"state": {
|
| 5 |
+
"single_arm_eef_xyz": {
|
| 6 |
+
"max": [
|
| 7 |
+
44.82416915893555,
|
| 8 |
+
74.993896484375,
|
| 9 |
+
7.081694602966309
|
| 10 |
+
],
|
| 11 |
+
"min": [
|
| 12 |
+
-57.89995193481445,
|
| 13 |
+
-10.476190567016602,
|
| 14 |
+
-68.30281066894531
|
| 15 |
+
],
|
| 16 |
+
"mean": [
|
| 17 |
+
-7.281844013035281,
|
| 18 |
+
29.96592333167826,
|
| 19 |
+
-25.618637384699355
|
| 20 |
+
],
|
| 21 |
+
"std": [
|
| 22 |
+
23.979941732549644,
|
| 23 |
+
24.416230415004133,
|
| 24 |
+
19.43252338277194
|
| 25 |
+
],
|
| 26 |
+
"q01": [
|
| 27 |
+
-56.414066314697266,
|
| 28 |
+
-7.399267196655273,
|
| 29 |
+
-60.488399505615234
|
| 30 |
+
],
|
| 31 |
+
"q99": [
|
| 32 |
+
44.62605285644531,
|
| 33 |
+
68.88888549804688,
|
| 34 |
+
5.503016471862793
|
| 35 |
+
]
|
| 36 |
+
},
|
| 37 |
+
"single_arm_eef_rpy": {
|
| 38 |
+
"max": [
|
| 39 |
+
3.1413190364837646,
|
| 40 |
+
2.808302879333496,
|
| 41 |
+
0.547160267829895
|
| 42 |
+
],
|
| 43 |
+
"min": [
|
| 44 |
+
-15.91375732421875,
|
| 45 |
+
-0.42427751421928406,
|
| 46 |
+
-100.0
|
| 47 |
+
],
|
| 48 |
+
"mean": [
|
| 49 |
+
-12.876365340030208,
|
| 50 |
+
2.1934915151454386,
|
| 51 |
+
-89.06816732241624
|
| 52 |
+
],
|
| 53 |
+
"std": [
|
| 54 |
+
4.438364737648521,
|
| 55 |
+
0.7593419392130816,
|
| 56 |
+
30.92162555839885
|
| 57 |
+
],
|
| 58 |
+
"q01": [
|
| 59 |
+
-14.98973274230957,
|
| 60 |
+
-0.34675517678260803,
|
| 61 |
+
-100.0
|
| 62 |
+
],
|
| 63 |
+
"q99": [
|
| 64 |
+
3.1062824726104736,
|
| 65 |
+
2.759462833404541,
|
| 66 |
+
0.455392986536026
|
| 67 |
+
]
|
| 68 |
+
},
|
| 69 |
+
"gripper": {
|
| 70 |
+
"max": [
|
| 71 |
+
31.742507934570312
|
| 72 |
+
],
|
| 73 |
+
"min": [
|
| 74 |
+
0.0
|
| 75 |
+
],
|
| 76 |
+
"mean": [
|
| 77 |
+
8.453739841643333
|
| 78 |
+
],
|
| 79 |
+
"std": [
|
| 80 |
+
7.863242578422885
|
| 81 |
+
],
|
| 82 |
+
"q01": [
|
| 83 |
+
0.0
|
| 84 |
+
],
|
| 85 |
+
"q99": [
|
| 86 |
+
27.7469482421875
|
| 87 |
+
]
|
| 88 |
+
}
|
| 89 |
+
},
|
| 90 |
+
"action": {
|
| 91 |
+
"single_arm_eef_xyz": {
|
| 92 |
+
"max": [
|
| 93 |
+
45.03376388549805,
|
| 94 |
+
73.33577728271484,
|
| 95 |
+
77.4813461303711
|
| 96 |
+
],
|
| 97 |
+
"min": [
|
| 98 |
+
-60.38243865966797,
|
| 99 |
+
-10.664224624633789,
|
| 100 |
+
-71.25431060791016
|
| 101 |
+
],
|
| 102 |
+
"mean": [
|
| 103 |
+
-7.273999378171898,
|
| 104 |
+
28.893785842769137,
|
| 105 |
+
-25.522589628544708
|
| 106 |
+
],
|
| 107 |
+
"std": [
|
| 108 |
+
24.13997737859988,
|
| 109 |
+
23.95556826391448,
|
| 110 |
+
23.182489643655508
|
| 111 |
+
],
|
| 112 |
+
"q01": [
|
| 113 |
+
-56.567874908447266,
|
| 114 |
+
-8.664224624633789,
|
| 115 |
+
-62.96855163574219
|
| 116 |
+
],
|
| 117 |
+
"q99": [
|
| 118 |
+
45.03376388549805,
|
| 119 |
+
67.33577728271484,
|
| 120 |
+
32.89389419555664
|
| 121 |
+
]
|
| 122 |
+
},
|
| 123 |
+
"single_arm_eef_rpy": {
|
| 124 |
+
"max": [
|
| 125 |
+
3.1413190364837646,
|
| 126 |
+
2.612942695617676,
|
| 127 |
+
0.547160267829895
|
| 128 |
+
],
|
| 129 |
+
"min": [
|
| 130 |
+
-14.98973274230957,
|
| 131 |
+
-0.42427751421928406,
|
| 132 |
+
-100.0
|
| 133 |
+
],
|
| 134 |
+
"mean": [
|
| 135 |
+
-13.402412390908477,
|
| 136 |
+
2.3359100808290307,
|
| 137 |
+
-89.40878290362456
|
| 138 |
+
],
|
| 139 |
+
"std": [
|
| 140 |
+
4.616244254847214,
|
| 141 |
+
0.8055287371739884,
|
| 142 |
+
30.77510217813324
|
| 143 |
+
],
|
| 144 |
+
"q01": [
|
| 145 |
+
-14.98973274230957,
|
| 146 |
+
-0.35010501742362976,
|
| 147 |
+
-100.0
|
| 148 |
+
],
|
| 149 |
+
"q99": [
|
| 150 |
+
3.107100009918213,
|
| 151 |
+
2.612942695617676,
|
| 152 |
+
0.455392986536026
|
| 153 |
+
]
|
| 154 |
+
},
|
| 155 |
+
"gripper": {
|
| 156 |
+
"max": [
|
| 157 |
+
32.0
|
| 158 |
+
],
|
| 159 |
+
"min": [
|
| 160 |
+
0.0
|
| 161 |
+
],
|
| 162 |
+
"mean": [
|
| 163 |
+
5.719525878968183
|
| 164 |
+
],
|
| 165 |
+
"std": [
|
| 166 |
+
8.653996030874026
|
| 167 |
+
],
|
| 168 |
+
"q01": [
|
| 169 |
+
0.0
|
| 170 |
+
],
|
| 171 |
+
"q99": [
|
| 172 |
+
28.0
|
| 173 |
+
]
|
| 174 |
+
}
|
| 175 |
+
}
|
| 176 |
+
},
|
| 177 |
+
"modalities": {
|
| 178 |
+
"video": {
|
| 179 |
+
"webcam": {
|
| 180 |
+
"resolution": [
|
| 181 |
+
640,
|
| 182 |
+
480
|
| 183 |
+
],
|
| 184 |
+
"channels": 3,
|
| 185 |
+
"fps": 30.0
|
| 186 |
+
}
|
| 187 |
+
},
|
| 188 |
+
"state": {
|
| 189 |
+
"single_arm_eef_xyz": {
|
| 190 |
+
"absolute": true,
|
| 191 |
+
"rotation_type": null,
|
| 192 |
+
"shape": [
|
| 193 |
+
3
|
| 194 |
+
],
|
| 195 |
+
"continuous": true
|
| 196 |
+
},
|
| 197 |
+
"single_arm_eef_rpy": {
|
| 198 |
+
"absolute": true,
|
| 199 |
+
"rotation_type": null,
|
| 200 |
+
"shape": [
|
| 201 |
+
3
|
| 202 |
+
],
|
| 203 |
+
"continuous": true
|
| 204 |
+
},
|
| 205 |
+
"gripper": {
|
| 206 |
+
"absolute": true,
|
| 207 |
+
"rotation_type": null,
|
| 208 |
+
"shape": [
|
| 209 |
+
1
|
| 210 |
+
],
|
| 211 |
+
"continuous": true
|
| 212 |
+
}
|
| 213 |
+
},
|
| 214 |
+
"action": {
|
| 215 |
+
"single_arm_eef_xyz": {
|
| 216 |
+
"absolute": true,
|
| 217 |
+
"rotation_type": null,
|
| 218 |
+
"shape": [
|
| 219 |
+
3
|
| 220 |
+
],
|
| 221 |
+
"continuous": true
|
| 222 |
+
},
|
| 223 |
+
"single_arm_eef_rpy": {
|
| 224 |
+
"absolute": true,
|
| 225 |
+
"rotation_type": null,
|
| 226 |
+
"shape": [
|
| 227 |
+
3
|
| 228 |
+
],
|
| 229 |
+
"continuous": true
|
| 230 |
+
},
|
| 231 |
+
"gripper": {
|
| 232 |
+
"absolute": true,
|
| 233 |
+
"rotation_type": null,
|
| 234 |
+
"shape": [
|
| 235 |
+
1
|
| 236 |
+
],
|
| 237 |
+
"continuous": true
|
| 238 |
+
}
|
| 239 |
+
}
|
| 240 |
+
},
|
| 241 |
+
"embodiment_tag": "new_embodiment"
|
| 242 |
+
}
|
| 243 |
+
}
|
model-00001-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b01a948ae542e4f9f1ff096165c1c2767aba566694cc92510e7335ce6c7ff07b
|
| 3 |
+
size 4999367032
|
model-00002-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ff4379eecf11fdd50745bd3acac9295d707b1aa46415aacd1df938c07bcb9875
|
| 3 |
+
size 2586705312
|
model.safetensors.index.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|