Rationale Support Quality and Alignment on OpenBookQA
97.58Rationale Quality (Q)SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFTModel=Ministral 3 8B2026.06 | 97.58 | 2.58 | |
| CoRAModel=Ministral 3 8B2026.06 | 96.09 | 3.66 | |
| GRPOModel=Ministral 3 8B2026.06 | 95.91 | 4.59 | |
| BaseModel=Ministral 3 8B2026.06 | 95.34 | 5.28 | |
| CoRAModel=Qwen2.5 7B2026.06 | 94.65 | 5.65 | |
| CoRAModel=Gemma 2 9B2026.06 | 92.6 | 7.29 | |
| GRPOModel=Gemma 2 9B2026.06 | 92.14 | 7.59 | |
| BaseModel=Qwen2.5 7B2026.06 | 91.94 | 10.44 | |
| SFTModel=Qwen2.5 7B2026.06 | 91.92 | 8.48 | |
| GRPOModel=Qwen2.5 7B2026.06 | 90.9 | 9.46 | |
| SFTModel=Gemma 2 9B2026.06 | 90.54 | 9.06 | |
| BaseModel=Gemma 2 9B2026.06 | 90.38 | 9.02 |