All requests submitted at once — engine queues them, processes B=1 at a time. E2E includes queue wait.
| Engine | B | req/s | tok/s | E2E p50 | E2E p99 |
|---|---|---|---|---|---|
| liteinfer | 1 | 1.75base | 72.61base | 11607.0 msbase | 18248.8 msbase |
| liteinfer-kvcache | 1 | 1.811.03x | 72.471.00x | 10190.2 ms1.14x | 17647.9 ms1.03x |
| liteinfer-native-kvcache | 1 | 1.811.03x | 72.160.99x | 10232.8 ms1.13x | 17724.8 ms1.03x |
| liteinfer-paged-kvcache | 1 | 1.570.89x | 62.700.86x | 11771.8 ms0.99x | 20397.1 ms0.89x |
| liteinfer-b4 | 4 | 4.312.46x | 177.062.44x | 4220.3 ms2.75x | 7420.9 ms2.46x |
| vllm | 1 | 2.901.65x | 180.712.49x | 5795.9 ms2.00x | 10998.8 ms1.66x |
| vllm-b4 | 4 | 10.776.14x | 655.899.03x | 1625.3 ms7.14x | 2931.8 ms6.22x |
Sequential, no queue — each request sent only after previous finishes. Pure per-request engine latency.
| Engine | B | TTFT p50 | TTFT p99 | E2E p50 | tok/s |
|---|---|---|---|---|---|
| liteinfer | 1 | 13.7 msbase | 15.4 msbase | 1710.4 msbase | 72.37base |
| liteinfer-kvcache | 1 | 14.6 ms0.94x | 16.8 ms0.92x | 1540.8 ms1.11x | 72.641.00x |
| liteinfer-native-kvcache | 1 | 13.9 ms0.99x | 15.6 ms0.99x | 1539.5 ms1.11x | 72.651.00x |
| liteinfer-paged-kvcache | 1 | 14.7 ms0.94x | 16.7 ms0.92x | 1829.0 ms0.94x | 61.240.85x |
| vllm | 1 | 25.9 ms0.53x | 28.8 ms0.54x | 693.1 ms2.47x | 182.432.52x |