Science Question Answering on QASCr (val)
79.4AccuracyIA2 → SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IA2 → SFTModel=Qwen3-4B-Base, N=4, Source Dataset=SciQr, OOD=true2025.09 | 79.4 | 0.09 | |
| SFT onlyModel=Qwen3-4B-Base, N=4, Source Dataset=SciQr, OOD=true2025.09 | 76.5 | 0.09 | |
| IA2 onlyModel=Qwen3-4B-Base, N=4, Source Dataset=SciQr, OOD=true2025.09 | 71.2 | 0.12 | |
| ICLModel=Qwen3-4B-Base, N=4, Source Dataset=SciQr, OOD=true2025.09 | 56.5 | 0.08 |