Long-Context Inference on LongBench <32K tokens v2 (short)
41.7AccuracyH2O (Best)
Evaluation results
| Method | Links | |
|---|---|---|
| H2O (Best)KV Cache Budget=20%2026.02 | 41.7 | |
| CHESS (Moderate)KV Cache Budget=40%2026.02 | 41.7 | |
| Quest (Best)KV Cache Budget=2048 tokens2026.02 | 40 | |
| CHESS (Aggressive)KV Cache Budget=1%2026.02 | 40 | |
| FullKV InferenceKV Cache Budget=100%2026.02 | 38.3 | |
| CHESS (Conservative)KV Cache Budget=73%2026.02 | 36.1 | |
| SnapKV (Best)KV Cache Budget=4096 tokens2026.02 | 34.4 | |
| KeyDiff (Best)KV Cache Budget=8192 tokens2026.02 | 30.6 |