Add robotics pipeline tag and link paper, project page, and GitHub repository
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,6 +1,7 @@
|
|
| 1 |
---
|
| 2 |
-
license: other
|
| 3 |
library_name: jax
|
|
|
|
|
|
|
| 4 |
tags:
|
| 5 |
- warp-rm
|
| 6 |
- openpi
|
|
@@ -11,25 +12,28 @@ tags:
|
|
| 11 |
|
| 12 |
# WARP-RM paper simulation policy checkpoints
|
| 13 |
|
| 14 |
-
This repository contains the two model-only policy parameter trees used for
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
training checkpoint.
|
| 18 |
|
| 19 |
-
The released directories are complete serving parameter trees when used with
|
| 20 |
-
the public OpenPI code. Their redistribution and use remain subject to the
|
| 21 |
-
applicable upstream OpenPI/Pi0 terms; this repository therefore uses the
|
| 22 |
-
custom license setting rather than asserting an Apache-2.0 grant for
|
| 23 |
-
upstream-derived model weights.
|
| 24 |
|
| 25 |
Layout:
|
| 26 |
|
| 27 |
* `vanilla/params/` and `vanilla/assets/` are the baseline policy tree.
|
| 28 |
-
* `warp_rabc_sss15/params/` and `warp_rabc_sss15/assets/` are the WARP-RM
|
| 29 |
-
reward-aligned BC policy tree.
|
| 30 |
|
| 31 |
-
Pass either containing directory as the local policy checkpoint directory to
|
| 32 |
-
the released OpenPI serving command. The `assets/` directory is required; do
|
| 33 |
-
not rename or discard it.
|
| 34 |
|
| 35 |
Files and SHA256 checksums are listed in `MANIFEST.json`.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
|
|
|
| 2 |
library_name: jax
|
| 3 |
+
license: other
|
| 4 |
+
pipeline_tag: robotics
|
| 5 |
tags:
|
| 6 |
- warp-rm
|
| 7 |
- openpi
|
|
|
|
| 12 |
|
| 13 |
# WARP-RM paper simulation policy checkpoints
|
| 14 |
|
| 15 |
+
This repository contains the two model-only policy parameter trees used for the paper [WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation](https://huggingface.co/papers/2606.28320): the baseline policy and the WARP-RM RABC policy. It omits optimizer/train state and does not contain a redistributed training checkpoint.
|
| 16 |
+
|
| 17 |
+
[Project Page](https://uynitsuj.github.io/warp-rm/) | [GitHub Repository](https://github.com/uynitsuj/WARP-RM)
|
|
|
|
| 18 |
|
| 19 |
+
The released directories are complete serving parameter trees when used with the public OpenPI code. Their redistribution and use remain subject to the applicable upstream OpenPI/Pi0 terms; this repository therefore uses the custom license setting rather than asserting an Apache-2.0 grant for upstream-derived model weights.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
Layout:
|
| 22 |
|
| 23 |
* `vanilla/params/` and `vanilla/assets/` are the baseline policy tree.
|
| 24 |
+
* `warp_rabc_sss15/params/` and `warp_rabc_sss15/assets/` are the WARP-RM reward-aligned BC policy tree.
|
|
|
|
| 25 |
|
| 26 |
+
Pass either containing directory as the local policy checkpoint directory to the released OpenPI serving command. The `assets/` directory is required; do not rename or discard it.
|
|
|
|
|
|
|
| 27 |
|
| 28 |
Files and SHA256 checksums are listed in `MANIFEST.json`.
|
| 29 |
+
|
| 30 |
+
## Citation
|
| 31 |
+
|
| 32 |
+
```bibtex
|
| 33 |
+
@article{yu2026warp,
|
| 34 |
+
title={WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation},
|
| 35 |
+
author={Yu, Justin and Goldberg, Andrew and Kondap, Kavish and El-Refai, Karim and Ransing, Ethan and Chen, Qianzhong and Schwager, Mac and Shentu, Fred and Wu, Philipp and Goldberg, Ken},
|
| 36 |
+
journal={arXiv preprint arXiv:2606.28320},
|
| 37 |
+
year={2026}
|
| 38 |
+
}
|
| 39 |
+
```
|