Machine Translation on English-to-Kachin
27.15chrFOur RL
Evaluation Results
| Method | Links | |
|---|---|---|
| Our RLBase Model=Qwen3-4B-Base, Training Method=RL, Context=full2026.06 | 27.15 | |
| Our RLBase Model=Llama-3.2-3B-Instruct, Training Method=RL, Context=full2026.06 | 24.93 | |
| Our SFTBase Model=Qwen3-4B-Base, Training Method=SFT, Context=none2026.06 | 18.27 | |
| Qwen3-4B-BaseBase Model=Qwen3-4B-Base, Training Method=Base (untuned), Context=full2026.06 | 17.74 | |
| Llama-3.2-3B-InstBase Model=Llama-3.2-3B-Instruct, Training Method=Instruct-tuned, Context=full2026.06 | 15.45 | |
| Qwen3-4B-BaseBase Model=Qwen3-4B-Base, Training Method=Base (untuned), Context=none2026.06 | 11.95 | |
| Our RLBase Model=Qwen3-4B-Base, Training Method=RL, Context=none2026.06 | 7.52 | |
| Our RLBase Model=Llama-3.2-3B-Instruct, Training Method=RL, Context=none2026.06 | 4.3 | |
| Our SFTBase Model=Llama-3.2-3B-Instruct, Training Method=SFT, Context=full2026.06 | 2.69 | |
| Llama-3.2-3B-InstBase Model=Llama-3.2-3B-Instruct, Training Method=Instruct-tuned, Context=none2026.06 | 2.36 | |
| Our SFTBase Model=Llama-3.2-3B-Instruct, Training Method=SFT, Context=none2026.06 | 1.52 | |
| Our SFTBase Model=Qwen3-4B-Base, Training Method=SFT, Context=full2026.06 | 1.28 |