Long-Context Reasoning on InfiniteBench
36.51Overall ScoreFull KV
Evaluation Results
| Method | Links | |
|---|---|---|
| Full KVBackbone=Qwen2.5-14B, KV Cache Budget=Full2026.06 | 36.51 | |
| SEKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 34.83 | |
| SentenceKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 32.17 | |
| SemantiCacheBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 31.42 | |
| ChunkKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 30.84 | |
| PyramidKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 29.37 | |
| Full KVBackbone=Llama-3.1-8B, KV Cache Budget=Full2026.06 | 28.88 | |
| SnapKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 28.14 | |
| SEKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 27.94 | |
| SentenceKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 26.07 | |
| H2OBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 25.38 | |
| SemantiCacheBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 25.33 | |
| Full KVBackbone=Llama-3-8B, KV Cache Budget=Full2026.06 | 25.27 | |
| SEKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 24.83 | |
| ChunkKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 24.71 | |
| PyramidKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 23.62 | |
| SentenceKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 23.14 | |
| Full KVBackbone=Mistral-7B, KV Cache Budget=Full2026.06 | 22.77 | |
| SemantiCacheBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 22.47 | |
| SnapKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 22.47 | |
| StreamingLLMBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.06 | 22.17 | |
| ChunkKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 21.83 | |
| Full KVBackbone=Llama-3.2-3B, KV Cache Budget=Full2026.06 | 21.16 | |
| SEKVBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 21.13 | |
| PyramidKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 20.81 | |
| H2OBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 20.14 | |
| SnapKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 19.73 | |
| SEKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 19.72 | |
| SentenceKVBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 19.47 | |
| SemantiCacheBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 18.84 | |
| SentenceKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 18.34 | |
| ChunkKVBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 18.23 | |
| SemantiCacheBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 17.83 | |
| PyramidKVBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 17.44 | |
| H2OBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 17.43 | |
| StreamingLLMBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.06 | 17.31 | |
| ChunkKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 17.12 | |
| SnapKVBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 16.72 | |
| PyramidKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 16.43 | |
| SnapKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 15.84 | |
| H2OBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 14.83 | |
| StreamingLLMBackbone=Llama-3-8B, KV Cache Budget=10%2026.06 | 14.82 | |
| H2OBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 13.72 | |
| StreamingLLMBackbone=Mistral-7B, KV Cache Budget=10%2026.06 | 12.64 | |
| StreamingLLMBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.06 | 11.43 |