A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 19:30 UTC
[bench] add start_times field to vllm bench serve json result (#32667)
Signed-off-by: Kebe <mail@kebe7jun.com>
Kebe committed
Jan 22, 2026 at 16:10 UTC
1bf1a34b1903189fd5da2f41313d3b8085c8ce48
1 file changed
+2
vllm/benchmarks/serve.py
+2
@@ -958,6 +958,7 @@ async def benchmark(
958
"output_lens": actual_output_lens,
959
"ttfts": [output.ttft for output in outputs],
960
"itls": [output.itl for output in outputs],
961
+ "start_times": [output.start_time for output in outputs],
962
"generated_texts": [output.generated_text for output in outputs],
963
"errors": [output.error for output in outputs],
964
"max_output_tokens_per_s": metrics.max_output_tokens_per_s,
@@ -1721,6 +1722,7 @@ async def main_async(args: argparse.Namespace) -> dict[str, Any]:
1722
for field in [
1723
"input_lens",
1724
"output_lens",
1725
+ "start_times",
1726
"ttfts",
1727
"itls",
1728
"generated_texts",