Science Question Answering on GPQA Diamond (avg@k)
58.33Accuracy (avg@k)TRACE (FKL on Ky)
Evaluation Results
| Method | Links | |
|---|---|---|
| TRACE (FKL on Ky)Backbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 58.33 | |
| BaseBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 58.27 | |
| RLSDBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 57.26 | |
| TRACE (RKL on Ey)Backbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 56.44 | |
| GRPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 53.85 | |
| SDPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 39.9 | |
| RLSDBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 38.89 | |
| TRACE (RKL on Ey)Backbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 38.67 | |
| SRPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.05 | 37.31 | |
| BaseBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 37.18 | |
| GRPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 37.18 | |
| TRACE (FKL on Ky)Backbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 36.77 | |
| SRPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 35.82 | |
| SDPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.05 | 27.41 |