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 15, 08:30 UTC
[Bugfix] Disable prefix caching by default for benchmark (#18771)
Signed-off-by: cascade812 <cascade812@outlook.com>
cascade committed
May 27, 2025 at 17:18 UTC
51e98e4ffd696289bce1ec92c78cc7298dc1600b
1 file changed
+1
-1
vllm/benchmarks/latency.py
+1
-1
@@ -82,7 +82,7 @@ def add_cli_args(parser: argparse.ArgumentParser):
82
parser = EngineArgs.add_cli_args(parser)
83
# V1 enables prefix caching by default which skews the latency
84
# numbers. We need to disable prefix caching by default.
85
- parser.set_defaults(enable_prefix_caching=True)
85
+ parser.set_defaults(enable_prefix_caching=False)
86
87
88
def main(args: argparse.Namespace):