Single-Document Question Answering on LongBench
53.67MultifieldQA ScoreFullKV
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FullKVBackbone=Qwen3-8B, KV Cache Size=Full2026.03 | 53.67 | 36.87 | — | |
| DapQBackbone=Qwen3-8B, KV Cache Size=2562026.03 | 50.78 | 32.14 | — | |
| FullKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=Full2026.03 | 49.7 | 36.5 | — | |
| FullKVModel=Mistral-7B-Instruct-v0.2, KV cache size=Full2026.02 | 49.34 | 32.99 | 26.63 | |
| SnapKVBackbone=Qwen3-8B, KV Cache Size=2562026.03 | 49.29 | 32.4 | — | |
| SnapKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=2562026.03 | 47.87 | 31.12 | — | |
| PyramidKVBackbone=Qwen3-8B, KV Cache Size=2562026.03 | 47.81 | 30.41 | — | |
| DapQBackbone=Qwen3-8B, KV Cache Size=1282026.03 | 47.15 | 29.1 | — | |
| SnapKVBackbone=Qwen3-8B, KV Cache Size=1282026.03 | 46.43 | 29.41 | — | |
| PyramidKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=2562026.03 | 46.37 | 29.65 | — | |
| FullKVModel=Qwen2-7B-Instruct, KV cache size=Full2026.02 | 45.04 | 42.35 | 25.14 | |
| DapQBackbone=Qwen2.5-7B-Instruct, KV Cache Size=2562026.03 | 45 | 30.21 | — | |
| H2OBackbone=Qwen3-8B, KV Cache Size=2562026.03 | 44.92 | 27.8 | — | |
| CompilerKVModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 43.2 | 24.06 | 22.31 | |
| DapQBackbone=Qwen3-8B, KV Cache Size=642026.03 | 43.17 | 25.78 | — | |
| DapQBackbone=Qwen2.5-7B-Instruct, KV Cache Size=1282026.03 | 43.12 | 26.81 | — | |
| PyramidKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=1282026.03 | 43.09 | 26.37 | — | |
| DynamicKVModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 43.08 | 23.65 | 22.05 | |
| PyramidKVModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 43.04 | 22.52 | 22.11 | |
| SnapKVModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 42.98 | 21.28 | 20.11 | |
| H2OBackbone=Qwen2.5-7B-Instruct, KV Cache Size=2562026.03 | 42.94 | 27.76 | — | |
| CompilerKVModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 42.81 | 35.93 | 23.12 | |
| DapQBackbone=Qwen2.5-7B-Instruct, KV Cache Size=642026.03 | 42.65 | 25.58 | — | |
| DynamicKVModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 42.62 | 35.57 | 22.77 | |
| SnapKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=1282026.03 | 41.81 | 27.46 | — | |
| H2OBackbone=Qwen3-8B, KV Cache Size=1282026.03 | 41.37 | 26.6 | — | |
| SnapKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=642026.03 | 40.66 | 22.9 | — | |
| FullKVBackbone=Llama3-8B-Instruct, KV Cache Size=Full2026.03 | 40.64 | 37.72 | — | |
| PyramidKVBackbone=Qwen3-8B, KV Cache Size=1282026.03 | 40.48 | 26.22 | — | |
| PyramidKVBackbone=Qwen3-8B, KV Cache Size=642026.03 | 40.44 | 25.32 | — | |
| FullKVModel=InternLM-2.5-Chat-1M, KV cache size=Full2026.02 | 39.98 | 27.61 | 22.42 | |
| PyramidKVModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 39.95 | 31.7 | 20.5 | |
| SnapKVBackbone=Qwen3-8B, KV Cache Size=642026.03 | 39.89 | 25.09 | — | |
| FullKVModel=LLaMA-3-8B-Instruct, KV cache size=Full2026.02 | 39.59 | 31.81 | 25.16 | |
| SnapKVModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 38.95 | 31.62 | 22.26 | |
| H2OBackbone=Qwen3-8B, KV Cache Size=642026.03 | 38.94 | 25.55 | — | |
| H2OModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 38.6 | 21.64 | 21.66 | |
| SnapKVBackbone=Llama3-8B-Instruct, KV Cache Size=2562026.03 | 38.39 | 30.84 | — | |
| PyramidKVBackbone=Llama3-8B-Instruct, KV Cache Size=1282026.03 | 38.29 | 28.8 | — | |
| DapQBackbone=Llama3-8B-Instruct, KV Cache Size=2562026.03 | 38.18 | 32.55 | — | |
| PyramidKVBackbone=Llama3-8B-Instruct, KV Cache Size=2562026.03 | 38.11 | 30.88 | — | |
| SnapKVBackbone=Llama3-8B-Instruct, KV Cache Size=1282026.03 | 37.8 | 29.52 | — | |
| H2OBackbone=Qwen2.5-7B-Instruct, KV Cache Size=1282026.03 | 37.8 | 26.83 | — | |
| DapQBackbone=Llama3-8B-Instruct, KV Cache Size=642026.03 | 37.36 | 25.99 | — | |
| DapQBackbone=Llama3-8B-Instruct, KV Cache Size=1282026.03 | 37.24 | 28.76 | — | |
| H2OBackbone=Llama3-8B-Instruct, KV Cache Size=2562026.03 | 36.63 | 28.11 | — | |
| H2OBackbone=Llama3-8B-Instruct, KV Cache Size=1282026.03 | 36.25 | 25.95 | — | |
| SnapKVModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 35.29 | 17.44 | 19.65 | |
| PyramidKVBackbone=Qwen2.5-7B-Instruct, KV Cache Size=642026.03 | 35.19 | 22.36 | — | |
| CompilerKVModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 34.61 | 20.32 | 18.36 | |
| H2OModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 34.22 | 30.43 | 20.33 | |
| DynamicKVModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 34.15 | 19.89 | 17.93 | |
| PyramidKVModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 33.48 | 17.35 | 18.8 | |
| CompilerKVModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 33.34 | 15.22 | 22.46 | |
| DynamicKVModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 32.94 | 14.93 | 22.1 | |
| SnapKVBackbone=Llama3-8B-Instruct, KV Cache Size=642026.03 | 32.92 | 25.06 | — | |
| SnapKVModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 32.38 | 12.37 | 21.71 | |
| H2OBackbone=Qwen2.5-7B-Instruct, KV Cache Size=642026.03 | 32.36 | 24.33 | — | |
| PyramidKVBackbone=Llama3-8B-Instruct, KV Cache Size=642026.03 | 31.8 | 22.04 | — | |
| H2OBackbone=Llama3-8B-Instruct, KV Cache Size=642026.03 | 30.83 | 24.02 | — | |
| PyramidKVModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 30.73 | 16.65 | 22.26 | |
| StreamingLLMModel=Mistral-7B-Instruct-v0.2, KV cache size=1282026.02 | 30.36 | 14.76 | 16.58 | |
| H2OModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 28.49 | 12.54 | 21.58 | |
| H2OModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 27.94 | 17.71 | 16.16 | |
| StreamingLLMModel=Qwen2-7B-Instruct, KV cache size=1282026.02 | 26.51 | 23.63 | 19.25 | |
| StreamingLLMModel=InternLM-2.5-Chat-1M, KV cache size=1282026.02 | 24.31 | 13.02 | 17.91 | |
| StreamingLLMModel=LLaMA-3-8B-Instruct, KV cache size=1282026.02 | 23.09 | 9.5 | 17.85 | |
| LUCID-PaTH2026.02 | 0.149 | 11.7 | — | |
| LUCID2026.02 | 0.144 | 10.55 | — | |
| Diff2026.02 | 0.139 | 8.7 | — | |
| PaTH2026.02 | 0.139 | 10.57 | — | |
| GSA2026.02 | 0.116 | 7.54 | — | |
| GLA2026.02 | 0.114 | 8.17 | — | |
| Standard2026.02 | 0.113 | 7.69 | — | |
| DeltaNet2026.02 | 0.076 | 7.06 | — |