LLM Inference Benchmark Lab

Loading run

Reading data/local-cpu-smollm2-135m.json.

Loading summary…

Configuration

Field Value

TTFT by concurrency

Client time to first streamed token. p50 and p95, warmup excluded.

Throughput vs concurrency

Successful requests and output tokens, divided by the measurement window.

End-to-end latency

Histogram

CDF

Metric N Mean p50 p95 p99

Errors

Measured requests only. Warmup failures are listed in the methodology panel.

Request Level Prompt Type Message

Methodology