Instruction Following on AlpacaEval Swahili
83Win RateCLO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLOModel=Llama-3-8B, Eval Lang=Target2025.05 | 83 | 17.6 | |
| CLOModel=Qwen-2.5-3B, Eval Lang=Target2025.05 | 74.7 | 23 | |
| CLOModel=Llama-2-13B, Eval Lang=Target2025.05 | 70.9 | 17 | |
| SFT+DPOModel=Llama-3-8B, Eval Lang=Target2025.05 | 65.4 | — | |
| CLOModel=Llama-2-7B, Eval Lang=Target2025.05 | 65 | 0.9 | |
| CLOModel=Mistral-7B-v0.1, Eval Lang=English2025.05 | 64.2 | 13 | |
| SFT+DPOModel=Llama-2-7B, Eval Lang=Target2025.05 | 64.1 | — | |
| CLOModel=Llama-3-8B, Eval Lang=English2025.05 | 61.4 | 11.1 | |
| CLOModel=Qwen-2.5-3B, Eval Lang=English2025.05 | 58.1 | 7 | |
| CLOModel=Llama-2-7B, Eval Lang=English2025.05 | 55.5 | 4 | |
| CLOModel=Llama-2-13B, Eval Lang=English2025.05 | 55 | 4.2 | |
| SFT+DPOModel=Llama-2-13B, Eval Lang=Target2025.05 | 53.9 | — | |
| SFT+DPOModel=Qwen-2.5-3B, Eval Lang=Target2025.05 | 51.7 | — | |
| SFT+DPOModel=Llama-2-7B, Eval Lang=English2025.05 | 51.5 | — | |
| CLOModel=Mistral-7B-v0.1, Eval Lang=Target2025.05 | 51.3 | 15.9 | |
| SFT+DPOModel=Mistral-7B-v0.1, Eval Lang=English2025.05 | 51.2 | — | |
| SFT+DPOModel=Qwen-2.5-3B, Eval Lang=English2025.05 | 51.1 | — | |
| SFT+DPOModel=Llama-2-13B, Eval Lang=English2025.05 | 50.8 | — | |
| SFT+DPOModel=Llama-3-8B, Eval Lang=English2025.05 | 50.3 | — | |
| SFT+DPOModel=Mistral-7B-v0.1, Eval Lang=Target2025.05 | 35.4 | — |