Add private offboarding model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
library_name: jax
|
| 4 |
+
tags:
|
| 5 |
+
- robotics
|
| 6 |
+
- preference-optimization
|
| 7 |
+
- flow-dpo
|
| 8 |
+
- metarobobench
|
| 9 |
+
- pi05
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# MetaRoboBench Pi05 Flow-DPO Checkpoints
|
| 13 |
+
|
| 14 |
+
Private offboarding archive of complete Flow-DPO parameter trees and training metrics.
|
| 15 |
+
|
| 16 |
+
Available complete runs:
|
| 17 |
+
|
| 18 |
+
- `checkpoints/g16_s500_conservative`: completed step 500
|
| 19 |
+
- `checkpoints/g4_s3000_b01_a005`: completed step 3,000
|
| 20 |
+
- `checkpoints/g8_s1000_b01_a005_r2`: completed step 1,000
|
| 21 |
+
|
| 22 |
+
These runs were initialized from the frame-aligned Pi05 SFT family. They are research checkpoints, not the main fixed-speed performance baseline. Incomplete 20,000-step and pilot runs are excluded because no full parameter tree remains.
|
| 23 |
+
|
| 24 |
+
Code reference:
|
| 25 |
+
|
| 26 |
+
- Repository: https://github.com/MMMM3202/MetaRoboBench
|
| 27 |
+
- Branch: `codex/vlm-rainbow-shared-20260821`
|
| 28 |
+
- Commit: `5b36567f0c6174ad82a3d4c669134f7b28acc945`
|
| 29 |
+
|
| 30 |
+
See `MODEL_MANIFEST.md` for the full inventory and result caveats.
|