List your company

LLM Inference Efficiency on LongBench-E

3.032Prefill Time (sec)Exact
3.003083.198293.39353.58871Feb 11, 2025

Evaluation results

MethodLinks
Exact2025.023.03237.769
BalanceKVcompression rate=0.25, b=256, T=22025.023.66238.054
StreamingLLMcompression rate=0.252025.023.68140.276
PyramidKVcompression rate=0.252025.023.74837.241
SnapKVcompression rate=0.252025.023.75540.426