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 14, 00:30 UTC
[benchmark] Make request IDs unique across clients by default (#27723)
Signed-off-by: Seiji Eicher <seiji@anyscale.com>
Seiji Eicher committed
Oct 30, 2025 at 19:40 UTC
b2e65cb4a7ea7c000517a7b78a6e0ccd9ecb0517
1 file changed
+2
-1
vllm/benchmarks/serve.py
+2
-1
@@ -26,6 +26,7 @@ import os
26
import random
27
import shutil
28
import time
29
+import uuid
30
import warnings
31
from collections.abc import AsyncGenerator, Iterable
32
from dataclasses import dataclass
@@ -1160,7 +1161,7 @@ def add_cli_args(parser: argparse.ArgumentParser):
1161
"--request-id-prefix",
1162
type=str,
1163
required=False,
1163
- default="benchmark-serving",
1164
+ default=f"bench-{uuid.uuid4().hex[:8]}-",
1165
help="Specify the prefix of request id.",
1166
)
1167