Question Answering on ScienceQA (Mean confidence, OCG)
97.2Mean ConfidenceSDFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SDFTModel=Qwen3-8B, Framework=OPD2026.04 | 97.2 | 48.1 | |
| SDPOModel=Olmo-3-7B-Instruct, Framework=OPD2026.04 | 94.9 | 31.9 | |
| SDPOModel=Qwen3-8B, Framework=OPD2026.04 | 93.5 | 12.9 | |
| BaseModel=Qwen3-8B2026.04 | 89.7 | 58.7 | |
| SDFTModel=Olmo-3-7B-Instruct, Framework=OPD2026.04 | 88.2 | 43 | |
| CaOPDModel=Olmo-3-7B-Instruct, Distillation Strategy=SDPO2026.04 | 86.7 | 22.7 | |
| BaseModel=Olmo-3-7B-Instruct2026.04 | 83.1 | 49.3 | |
| CaOPDModel=Qwen3-8B, Distillation Strategy=SDPO2026.04 | 77 | -3.3 | |
| CaOPDModel=Olmo-3-7B-Instruct, Distillation Strategy=SDFT2026.04 | 52.9 | 8.1 | |
| CaOPDModel=Qwen3-8B, Distillation Strategy=SDFT2026.04 | 38.2 | -11.8 |