Long-context evaluation on RULER 256K
100NS1 (Sequence Accuracy 1)Dense
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DenseModel=Qwen2.5-14B-Instruct-1M, Sparse token budget=4096 (1.56%)2025.06 | 100 | 100 | 100 | 100 | 90 | 85 | 97.5 | 100 | 95 | 60 | 40 | 87.95 | — | |
| top-kModel=Qwen2.5-14B-Instruct-1M, Sparse token budget=4096 (1.56%)2025.06 | 100 | 100 | 100 | 100 | 90 | 81.25 | 98.75 | 100 | 88.33 | 60 | 40 | 87.12 | — | |
| HATAModel=Qwen2.5-14B-Instruct-1M, Sparse token budget=4096 (1.56%)2025.06 | 100 | 100 | 100 | 100 | 95 | 85 | 97.5 | 96 | 85 | 60 | 45 | 88.05 | — | |
| N-3-Super 120B-A12B-BaseShots=02026.04 | — | — | — | — | — | — | — | — | — | — | — | — | 84.56 |