Add checkpoint README
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: openpi
|
| 3 |
+
tags:
|
| 4 |
+
- openpi
|
| 5 |
+
- robotics
|
| 6 |
+
- jax
|
| 7 |
+
- orbax
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# laundry_only_white_0511
|
| 11 |
+
|
| 12 |
+
This repository contains OpenPI checkpoints uploaded from a local training run.
|
| 13 |
+
|
| 14 |
+
## Contents
|
| 15 |
+
|
| 16 |
+
- Source directory: `/home/jeong/openpi/checkpoints/pi05_daru_gripper/laundry_only_white_0511`
|
| 17 |
+
- Uploaded steps: `40000`, `50000`
|
| 18 |
+
- Includes `train_state`: `yes`
|
| 19 |
+
|
| 20 |
+
## Loading
|
| 21 |
+
|
| 22 |
+
Use the checkpoint directory for a specific step when creating a trained policy.
|
| 23 |
+
|
| 24 |
+
```python
|
| 25 |
+
from openpi.policies import policy_config as _policy_config
|
| 26 |
+
from openpi.training import config as _config
|
| 27 |
+
|
| 28 |
+
config = _config.get_config("YOUR_CONFIG_NAME")
|
| 29 |
+
policy = _policy_config.create_trained_policy(config, "PATH_OR_HF_SNAPSHOT_TO_STEP_DIR")
|
| 30 |
+
```
|
| 31 |
+
|
| 32 |
+
If you uploaded multiple steps, point OpenPI at the specific step subdirectory you want to use.
|