Skip to content

v1.0.11

Latest

Choose a tag to compare

@WarningRan WarningRan released this 11 Dec 22:40

infer/vllm/process: Capture full benchmark metrics in JSON output

Description: Updated process_server to dynamically parse the complete client benchmark table into a new metrics field. This standardizes keys (converting spaces to underscores, e.g., Median_TTFT_(ms)) and ensures forward compatibility for capturing all future vLLM performance indicators without code changes.