Long-context understanding on LongBench (Derived/Specific Metrics)
49.11FullKV PerformanceFullKV
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FullKVModel=Qwen3-8B, Average KV Budget=5122026.06 | 49.11 | — | — | — | — | |
| FullKVModel=LLaMA-3.1-8B, Average KV Budget=5122026.06 | 38.99 | — | — | — | — | |
| CAKEModel=LLaMA-3.1-8B, Average KV Budget=5122026.06 | — | 36.35 | — | — | — | |
| CAKEModel=Qwen3-8B, Average KV Budget=5122026.06 | — | 47.01 | — | — | — | |
| PolyKVModel=LLaMA-3.1-8B, Average KV Budget=5122026.06 | — | — | 37.79 | 37.59 | 1.44 | |
| PolyKVModel=Qwen3-8B, Average KV Budget=5122026.06 | — | — | 47.54 | 47.35 | 0.54 |