Mathematical Reasoning on MGSM average (test)
84.8AccuracyLanguage Steering
Evaluation Results
| Method | Links | |
|---|---|---|
| Language SteeringModel=Qwen2.5-14B-Instruct, Setting=Ours2026.02 | 84.8 | |
| OracleModel=Qwen2.5-14B-Instruct, Setting=OR2026.02 | 81.23 | |
| Multilingual Few-Shot BaselineModel=Qwen2.5-14B-Instruct, Setting=MFS2026.02 | 80.86 | |
| BaselineModel=Qwen2.5-14B-Instruct, Setting=Baseline (B)2026.02 | 79.4 | |
| OracleModel=Qwen2.5-7B-Instruct, Setting=OR2026.02 | 73.21 | |
| Language SteeringModel=Qwen2.5-7B-Instruct, Setting=Ours2026.02 | 71.83 | |
| Multilingual Few-Shot BaselineModel=Qwen2.5-7B-Instruct, Setting=MFS2026.02 | 69.94 | |
| BaselineModel=Qwen2.5-7B-Instruct, Setting=Baseline (B)2026.02 | 68.95 | |
| OracleModel=LLaMA-3.1-8B-Instruct, Setting=OR2026.02 | 68.16 | |
| Language SteeringModel=LLaMA-3.1-8B-Instruct, Setting=Ours2026.02 | 65.87 | |
| Multilingual Few-Shot BaselineModel=LLaMA-3.1-8B-Instruct, Setting=MFS2026.02 | 63.19 | |
| BaselineModel=LLaMA-3.1-8B-Instruct, Setting=Baseline (B)2026.02 | 61.01 |