Science Question Answering on Physics (mean@16, maj@16)
81.6Mean@16SC-SDPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SC-SDPOBackbone=Qwen3-8B, Alpha=0.52026.05 | 81.6 | 84.9 | |
| SDPOBackbone=Qwen3-8B, Filtering Strategy=Hard Filter2026.05 | 76.4 | 79.6 | |
| SDPOBackbone=Qwen3-8B2026.05 | 75.8 | 78.5 | |
| SC-SDPOBackbone=Qwen3-8B, Alpha=12026.05 | 75.4 | 77.3 | |
| SRPOBackbone=Qwen3-8B2026.05 | 73.4 | 76.3 | |
| SDPOBackbone=Qwen3-8B, Weighting Strategy=PACED2026.05 | 69.2 | 71.7 | |
| GRPOBackbone=Qwen3-8B, Normalization=None2026.05 | 69.1 | 76.5 | |
| GRPOBackbone=Qwen3-8B2026.05 | 64.7 | 68.3 | |
| Qwen3-8BBackbone=Qwen3-8B2026.05 | 57.7 | 53.8 |