Scientific Reasoning on GPQA Diamond (average@32)
55.19Average@32GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOParams=4B2026.05 | 55.19 | |
| SODParams=1.7B2026.05 | 38.72 | |
| OPSDgtParams=1.7B2026.05 | 35.02 | |
| GRPOParams=1.7B2026.05 | 33.55 | |
| OPSDhintParams=1.7B2026.05 | 33.46 | |
| OPDParams=1.7B2026.05 | 31.73 | |
| SFTParams=1.7B2026.05 | 29.85 | |
| VanillaParams=1.7B2026.05 | 26.8 | |
| SODParams=0.6B2026.05 | 22.19 | |
| GRPOParams=0.6B2026.05 | 20.38 | |
| OPDParams=0.6B2026.05 | 17.76 | |
| OPSDgtParams=0.6B2026.05 | 17.32 | |
| OPSDhintParams=0.6B2026.05 | 15.98 | |
| SFTParams=0.6B2026.05 | 15.2 | |
| VanillaParams=0.6B2026.05 | 13.24 |