Multi-task Language Understanding on MMLU (Score, Speedup)
69.5MMLU ScoreLLaMA-3-8B-Instruct
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaMA-3-8B-InstructData=ZsRE 10K2026.02 | 69.5 | — | |
| CrispEditData=ZsRE 10K2026.02 | 68.5 | — | |
| Full DecodingBackbone=Dream-7B2026.02 | 68.2 | 1 | |
| Adam-NSCLData=ZsRE 10K2026.02 | 68.2 | — | |
| LocBF-FTData=ZsRE 10K2026.02 | 68 | — | |
| UltraEditData=ZsRE 10K2026.02 | 67.9 | — | |
| Full DecodingBackbone=LLaDA-8B2026.02 | 67.3 | 1 | |
| JotBackbone=Dream-7B2026.02 | 66.7 | 1.57 | |
| JotBackbone=LLaDA-8B2026.02 | 64.5 | 1.98 | |
| KLASSBackbone=Dream-7B2026.02 | 64.4 | 1.06 | |
| KLASSBackbone=LLaDA-8B2026.02 | 63.4 | 1.26 | |
| ProphetBackbone=LLaDA-8B2026.02 | 63.3 | 2.08 | |
| ProphetBackbone=Dream-7B2026.02 | 60.2 | 1.4 | |
| AlphaEditData=ZsRE 10K2026.02 | 22.8 | — |