Long-context Question Answering on FRAMES C=12k
34.95AccuracyIntentKV-Phase-2
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| IntentKV-Phase-2Model=Qwen2.5-14B2026.06 | 34.95 | 85.71 | 16.6 | 16.7 | 18.7 | 13.9 | 76.8 | 7.14 | |
| Full-cacheModel=Qwen2.5-14B2026.06 | 34.71 | 86.6 | 23.9 | 24.3 | 39.6 | 39.9 | 236.4 | 8.86 | |
| H2OModel=Qwen2.5-14B2026.06 | 33.62 | 86.87 | 22.8 | 23.7 | 52.1 | 57 | 109.7 | 8.66 | |
| SnapKVModel=Qwen2.5-14B2026.06 | 33.25 | 86.63 | 23.8 | 24.2 | 39.9 | 40.3 | 86.8 | 8.82 | |
| StreamingLLMModel=Qwen2.5-14B2026.06 | 29.85 | 85.16 | 15.9 | 16.4 | 22.6 | 23 | 83.4 | 6.95 | |
| IntentKV-Phase-2Model=Qwen3-8B2026.06 | 29.25 | 82.11 | 12.8 | 12.9 | 6.8 | 6 | 28.9 | 7.53 | |
| SnapKVModel=Qwen3-8B2026.06 | 29 | 82.67 | 16.8 | 17 | 12.9 | 13 | 48.6 | 8.91 | |
| Full-cacheModel=Qwen3-8B2026.06 | 28.52 | 82.13 | 17.6 | 17.8 | 15.5 | 15.6 | 68.7 | 9.23 | |
| StreamingLLMModel=Qwen3-8B2026.06 | 27.91 | 82.48 | 15.4 | 15.6 | 8.3 | 8.4 | 33.6 | 8.5 | |
| H2OModel=Qwen3-8B2026.06 | 27.18 | 83 | 16.8 | 17 | 12.8 | 12.9 | 46 | 8.94 |