Translator-call mode selection on MMLU ProX Lite
76.45Macro F1LUAR
Evaluation Results
| Method | Links | |
|---|---|---|
| LUARBackbone=Qwen3-4B2026.06 | 76.45 | |
| LUARBackbone=Qwen3-8B2026.06 | 68.18 | |
| ST(qr)Backbone=Qwen3-4B2026.06 | 65.44 | |
| BOUNDARY-SFTBackbone=Qwen3-4B2026.06 | 65.21 | |
| ST(qr)Backbone=Qwen3-8B2026.06 | 61.13 | |
| NATIVE-TOOL-USEBackbone=Qwen3-4B2026.06 | 60.03 | |
| BOUNDARY-SFTBackbone=Qwen3-8B2026.06 | 59.47 | |
| ST(q)Backbone=Qwen3-4B2026.06 | 48.38 | |
| SELF-ASSESSMENTBackbone=Qwen3-8B2026.06 | 48.04 | |
| SELF-ASSESSMENTBackbone=Qwen3-4B2026.06 | 47.83 | |
| ST(q)Backbone=Qwen3-8B2026.06 | 46.8 | |
| NATIVE-TOOL-USEBackbone=Qwen3-8B2026.06 | 41.36 |