List your company

LLM Serving Latency on Crawler Workload 4.0 QPS, 10x delays

0.64P50 Latency ScorevLLM-NS
0.31482.50994.7056.9001Mar 29, 2026

Evaluation results

MethodLinks
vLLM-NSMode=Non-streaming baseline2026.030.648.97
MCPSEviction Strategy=Swap2026.033.860.48
MCPSEviction Strategy=Cost-Based2026.034.960.77
MCPSEviction Strategy=Recompute2026.035.920.73
vLLM-SEviction Strategy=Swap2026.037.290.66
LCASEviction Strategy=Swap2026.037.324.8
FCFSEviction Strategy=Swap2026.037.786.69
LCASEviction Strategy=Cost-Based2026.037.829.14
vLLM-SEviction Strategy=Cost-Based2026.038.250.71
FCFSEviction Strategy=Cost-Based2026.038.38.62
vLLM-SEviction Strategy=Recompute2026.038.590.78
LCASEviction Strategy=Recompute2026.038.619.23
FCFSEviction Strategy=Recompute2026.038.7710.03