Multiple-choice Question Answering on mCSQA
36.8Accuracy (ZH)ORACLE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ORACLEStrategy=Upper bound selection, Backbone=DeepSeek-v3.12026.04 | 36.8 | 45.7 | 39.9 | — | |
| ORACLEBase Model=Llama3.3-70B2026.04 | 33.4 | 41.8 | 36.2 | — | |
| CLASSIFIER (Ours)Strategy=Learned Routing (XGBoost), Backbone=DeepSeek-v3.12026.04 | 28.4 | 39.1 | 33.8 | — | |
| TRANSLATEStrategy=Translate-then-inference, Backbone=DeepSeek-v3.12026.04 | 28.2 | 38.5 | 33.4 | — | |
| NATIVEStrategy=Native inference, Backbone=DeepSeek-v3.12026.04 | 27.6 | 38.2 | 32.9 | — | |
| XGBOOST (Ours)Base Model=Llama3.3-70B2026.04 | 26.5 | 34.6 | 30.6 | — | |
| NATIVEBase Model=Llama3.3-70B2026.04 | 26.1 | 33.8 | 30 | — | |
| MLP (Ours)Base Model=Llama3.3-70B2026.04 | 26.1 | 35.2 | 30.6 | — | |
| TRANSLATEBase Model=Llama3.3-70B2026.04 | 24.3 | 33.5 | 28.9 | — |