Scientific Reasoning on Scientific Methods Reasoning Cross-Domain (test)
49.08AccuracyReason-SFT-DrGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Reason-SFT-DrGRPOModel=Qwen32026.04 | 49.08 | |
| Reason-SFT-DAPOModel=Qwen32026.04 | 48.37 | |
| Reason-DAPOModel=Qwen32026.04 | 45.96 | |
| Reasoning (high)Model=GPT-52026.04 | 45.96 | |
| Direct-SFTModel=Qwen32026.04 | 45.67 | |
| Reasoning (med)Model=GPT-52026.04 | 45.25 | |
| Reasoning (low)Model=GPT-52026.04 | 42.84 | |
| Synthetic-Reason-SFT-DAPOModel=Qwen32026.04 | 41.1 | |
| Direct-SFTModel=Llama3.12026.04 | 31.2 | |
| Reason-SFTModel=Qwen32026.04 | 29.31 | |
| Base (Reasoning)Model=Llama3.12026.04 | 18.22 | |
| Base (Reasoning)Model=Qwen32026.04 | 12.62 | |
| BaseModel=Llama3.12026.04 | 3.83 | |
| BaseModel=Qwen32026.04 | 3.55 |