Multiple-choice factual recall on POLYFACT Low
56.93AccuracyGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen-2.5-7B2026.06 | 56.93 | |
| GRPOBackbone=OLMo-2-1124-7B2026.06 | 54.48 | |
| CPT + GRPOBackbone=OLMo-2-1124-7B2026.06 | 54.41 | |
| BaselineBackbone=Qwen-2.5-7B2026.06 | 52.26 | |
| CPT + SFTBackbone=OLMo-2-1124-7B2026.06 | 52.02 | |
| CPTBackbone=OLMo-2-1124-7B2026.06 | 51.82 | |
| BaselineBackbone=OLMo-2-1124-7B2026.06 | 51.8 | |
| CPTBackbone=Qwen-2.5-7B2026.06 | 51.09 | |
| SFTBackbone=OLMo-2-1124-7B2026.06 | 50.04 | |
| CPT + GRPOBackbone=Qwen-2.5-7B2026.06 | 49.33 | |
| SFTBackbone=Qwen-2.5-7B2026.06 | 48.88 | |
| CPT + SFTBackbone=Qwen-2.5-7B2026.06 | 47.53 |