LLM Inference Performance on Context Length 10K
0.45Prefill Time (s)IndexCache
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| IndexCacheRetention Ratio=1/42026.03 | 0.45 | 91 | 3,310 | |
| IndexCacheRetention Ratio=1/22026.03 | 0.47 | 84.5 | 3,070 | |
| DSARetention Ratio=Full2026.03 | 0.57 | 73.5 | 2,700 |