Long-context Understanding on LongBench, LongBench v2 & LooGLE Combined
45.71Overall ScoreLongCrafter
Evaluation Results
| Method | Links | |
|---|---|---|
| LongCrafterBackbone=LLaMA-3.1-8B, SFT Method=LongCrafter2026.07 | 45.71 | |
| LongCrafterBackbone=Qwen2.5-7B, SFT Method=LongCrafter2026.07 | 45.15 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B, SFT Method=Instruct (official)2026.07 | 42.74 | |
| LLaMA-3.1-8B-InstructBackbone=LLaMA-3.1-8B, SFT Method=Instruct (official)2026.07 | 40.46 | |
| LongFaithBackbone=LLaMA-3.1-8B, SFT Method=LongFaith2026.07 | 40.41 | |
| LongMagpieBackbone=LLaMA-3.1-8B, SFT Method=LongMagpie2026.07 | 40.07 | |
| LongFaithBackbone=Qwen2.5-7B, SFT Method=LongFaith2026.07 | 39.35 | |
| LongMagpieBackbone=Qwen2.5-7B, SFT Method=LongMagpie2026.07 | 39.21 | |
| LongRewardBackbone=Qwen2.5-7B, SFT Method=LongReward2026.07 | 37.82 | |
| LongRewardBackbone=LLaMA-3.1-8B, SFT Method=LongReward2026.07 | 36.96 | |
| LongAlignBackbone=LLaMA-3.1-8B, SFT Method=LongAlign2026.07 | 36.83 | |
| LongAlignBackbone=Qwen2.5-7B, SFT Method=LongAlign2026.07 | 36.78 |