Machine Translation on WMT CS-EN (test)
21.6BLEUMix-MoE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Mix-MoEBase Model=Llama3.2-1B, Training Strategy=Mix-MoE2026.05 | 21.6 | 43.8 | 72.8 | |
| MOE-LPRBase Model=Llama3.2-1B, Training Strategy=MOE-LPR2026.05 | 20.9 | 42.5 | 72.1 | |
| MoEBase Model=Llama3.2-1B, Training Strategy=MoE2026.05 | 18.4 | 39.3 | 70.9 | |
| Full FinetuneBase Model=Llama3.2-1B, Training Strategy=Full Finetune2026.05 | 17.7 | 38.2 | 68.1 | |
| Qwen2.5-1.5BBase Model=Qwen2.5-1.5B2026.05 | 17.4 | 38.6 | 71.7 | |
| Dense PT+SFTBase Model=Llama3.2-1B, Training Strategy=Dense PT+SFT2026.05 | 16.2 | 40 | 69.5 | |
| LLAMA-MOE v2Base Model=Llama3.2-1B, Training Strategy=LLAMA-MOE v22026.05 | 13.8 | 30.9 | 60.7 | |
| LoRABase Model=Llama3.2-1B, Training Strategy=LoRA2026.05 | 12.7 | 30.8 | 62.6 | |
| LLAMA-ProBase Model=Llama3.2-1B, Training Strategy=LLAMA-Pro2026.05 | 11.9 | 30.9 | 59.3 | |
| Qwen2.5-0.5BBase Model=Qwen2.5-0.5B2026.05 | 10.6 | 27 | 64.5 | |
| Llama3.2-1BBase Model=Llama3.2-1B2026.05 | 9.7 | 22.7 | 59.3 |