Update model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: lerobot
|
| 3 |
+
tags:
|
| 4 |
+
- molmoact2
|
| 5 |
+
- robotics
|
| 6 |
+
- lerobot
|
| 7 |
+
- vla
|
| 8 |
+
|
| 9 |
+
base_model: allenai/MolmoAct2
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# molmoact2_tool_insert_quantile_norm_fix_25k
|
| 13 |
+
|
| 14 |
+
Fine-tuned [MolmoAct2](https://huggingface.co/allenai/MolmoAct2) (action-expert-only) for `tool_insert` on SO101 data.
|
| 15 |
+
|
| 16 |
+
| | |
|
| 17 |
+
|---|---|
|
| 18 |
+
| **Policy** | MolmoAct2 (`policy.type=molmoact2`) |
|
| 19 |
+
| **Init checkpoint** | [allenai/MolmoAct2](https://huggingface.co/allenai/MolmoAct2) |
|
| 20 |
+
| **Dataset** | [villekuosmanen/armnetbench_tool_insert](https://huggingface.co/datasets/villekuosmanen/armnetbench_tool_insert) |
|
| 21 |
+
| **Task** | `tool_insert` |
|
| 22 |
+
| **Action dim** | 6 (single-arm) |
|
| 23 |
+
| **Cameras** | `top`, `wrist`, `front` |
|
| 24 |
+
| **Training** | 25k steps, QUANTILES norm, freeze, batch 32 global, Isambard GH200 |
|
| 25 |
+
| **Prior HF repo** | [pravsels/molmoact2_tool_insert](https://huggingface.co/pravsels/molmoact2_tool_insert) |
|
| 26 |
+
| **W&B project** | [molmoact2_tool_insert_quantile_norm_fix_25k](https://wandb.ai/pravsels/molmoact2_tool_insert_quantile_norm_fix_25k) |
|
| 27 |
+
| **W&B run** | [oiew4qmt](https://wandb.ai/pravsels/molmoact2_tool_insert_quantile_norm_fix_25k/runs/oiew4qmt) |
|
| 28 |
+
|
| 29 |
+
## Checkpoints
|
| 30 |
+
|
| 31 |
+
The checkpoint (local step `025000`, 25k training steps) lives at the **repository root** for direct loading.
|
| 32 |
+
|
| 33 |
+
|
| 34 |
+
## Verification
|
| 35 |
+
|
| 36 |
+
| **Checkpoint step** | `025000` |
|
| 37 |
+
| **Source path** | `checkpoints/025000/pretrained_model/` |
|
| 38 |
+
| **model.safetensors** | 10,884,573,720 bytes, sha256 `3535baf761b841961728d93b6f5ab6895b9057cf04c3846989040baa2e517782` |
|
| 39 |
+
| **policy_preprocessor.json** | 2,495 bytes, sha256 `e7c8b8293cb0265a01f83278033272efcb21b4c7bfb031cfbd683ed74ee7b139` |
|
| 40 |
+
| **policy_postprocessor.json** | 757 bytes, sha256 `6dbed1e1ec69e8c50f3a04c1f144a54231e3ef508f15fd7896ead43ea645b033` |
|
| 41 |
+
| **train_config.json** | 8,257 bytes, sha256 `346d840c7769cee2cf86ec30041a738c02c95aea24b4e983903b9de172b0904c` |
|
| 42 |
+
|
| 43 |
+
Verify after download:
|
| 44 |
+
|
| 45 |
+
```bash
|
| 46 |
+
sha256sum model.safetensors
|
| 47 |
+
# expected: 3535baf761b841961728d93b6f5ab6895b9057cf04c3846989040baa2e517782
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
## Usage
|
| 51 |
+
|
| 52 |
+
```python
|
| 53 |
+
from lerobot.policies.molmoact2.modeling_molmoact2 import MolmoAct2Policy
|
| 54 |
+
policy = MolmoAct2Policy.from_pretrained("pravsels/molmoact2_tool_insert_quantile_norm_fix_25k")
|
| 55 |
+
```
|