English-to-Chinese Translation on WMT24
86.16GRFReflectMT-Stage2(Our)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ReflectMT-Stage2(Our)Token=47.572026.04 | 86.16 | 2.1 | 79.86 | |
| Qwen3-8B(w/ refl)Token=786.63, reflection=true2026.04 | 85.92 | 2.51 | 79.66 | |
| DeepSeek-R1(w/ refl)Token=1069.07, reflection=true2026.04 | 85.61 | 2.56 | 79.28 | |
| Qwen2.5-14B(w/ refl)Token=208.86, reflection=true2026.04 | 85.33 | 2.63 | 79.29 | |
| QwQ-32B(w/ refl)Token=1102, reflection=true2026.04 | 84.47 | 2.98 | 77.61 | |
| DeepSeek-R1Token=605.422026.04 | 84.03 | 2.64 | 77.93 | |
| ReflectMT-Stage1(Our)Token=41.392026.04 | 83.39 | 2.49 | 79.41 | |
| Qwen3-8BToken=654.062026.04 | 82.56 | 2.59 | 79.58 | |
| Marco-o1-7B(w/ refl)Token=763.97, reflection=true2026.04 | 81.88 | 2.88 | 78.18 | |
| ExTrans-7BToken=1957.792026.04 | 81.7 | 2.72 | 77.51 | |
| ReflectMT-SFT(Cold Start)Token=37.572026.04 | 80.82 | 2.73 | 79.12 | |
| Qwen-7B-Full-RL‡Token=245.99, thought_mode=full2026.04 | 79.92 | 2.16 | 79.67 | |
| Qwen-7B-Adapt-RL†Token=239.31, thought_mode=adaptive2026.04 | 79.4 | 2.15 | 79.68 | |
| QwQ-32BToken=684.292026.04 | 78.9 | 3.03 | 74.82 | |
| Qwen2.5-14BToken=43.382026.04 | 77.63 | 2.77 | 78.81 | |
| Qwen-7B-Adapt-LoRAToken=183.752026.04 | 76.59 | 2.57 | 78.74 | |
| DRT-7BToken=520.032026.04 | 76.39 | 3 | 77.4 | |
| DeepTrans-7BToken=2581.522026.04 | 76.22 | 2.99 | 77.39 | |
| Marco-o1-7BToken=584.32026.04 | 74.67 | 2.97 | 77.35 | |
| Qwen-7B-InstructToken=39.372026.04 | 69.51 | 3.23 | 76.51 | |
| Llama3.1-8BToken=52.342026.04 | 65.57 | 3.46 | 75.95 |