General Reasoning on GPQA (Avg@k, Pass@k)
35.66Avg@kSFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFTBackbone=Qwen2.5-7B-Instruct2026.05 | 35.66 | 65.63 | |
| IWBackbone=Qwen2.5-7B-Instruct2026.05 | 34.82 | 64.29 | |
| IWBackbone=Llama-3.1-8B-Instruct2026.05 | 34.32 | 69.2 | |
| RPSFTBackbone=Qwen2.5-7B-Instruct2026.05 | 34.21 | 65.18 | |
| RPSFTBackbone=Llama-3.1-8B-Instruct2026.05 | 33.48 | 68.08 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 32.76 | 59.82 | |
| SFTBackbone=Llama-3.1-8B-Instruct2026.05 | 32.7 | 65.4 | |
| DFTBackbone=Qwen2.5-7B-Instruct2026.05 | 30.47 | 58.04 | |
| IWBackbone=Qwen2.5-3B-Instruct2026.05 | 29.41 | 65.4 | |
| RPSFTBackbone=Qwen2.5-3B-Instruct2026.05 | 28.13 | 62.05 | |
| SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 27.73 | 61.61 | |
| BaseBackbone=Llama-3.1-8B-Instruct2026.05 | 27.62 | 61.38 | |
| BaseBackbone=Qwen2.5-3B-Instruct2026.05 | 27.18 | 54.46 | |
| DFTBackbone=Llama-3.1-8B-Instruct2026.05 | 26.79 | 57.14 | |
| DFTBackbone=Qwen2.5-3B-Instruct2026.05 | 26.45 | 55.13 |