Instruction Following on AlpacaEval LC (test)
84.8Win RateDeepSeek-V3
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek-V3Backbone=DeepSeek-V32026.02 | 84.8 | |
| RefEvalBackbone=Llama3-8B-Inst2026.02 | 73.1 | |
| RefEvalBackbone=Qwen2.5-7B2026.02 | 70 | |
| SimPOBackbone=Llama3-8B-Inst2026.02 | 51.6 | |
| Qwen2.5-7B-InstBackbone=Qwen2.5-7B2026.02 | 29.9 |