From 51e98e4ffd696289bce1ec92c78cc7298dc1600b Mon Sep 17 00:00:00 2001 From: cascade Date: Tue, 27 May 2025 17:18:09 -0700 Subject: [PATCH] [Bugfix] Disable prefix caching by default for benchmark (#18771) Signed-off-by: cascade812 --- vllm/benchmarks/latency.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/benchmarks/latency.py b/vllm/benchmarks/latency.py index 2c992727b139..0dd938e75129 100644 --- a/vllm/benchmarks/latency.py +++ b/vllm/benchmarks/latency.py @@ -82,7 +82,7 @@ def add_cli_args(parser: argparse.ArgumentParser): parser = EngineArgs.add_cli_args(parser) # V1 enables prefix caching by default which skews the latency # numbers. We need to disable prefix caching by default. - parser.set_defaults(enable_prefix_caching=True) + parser.set_defaults(enable_prefix_caching=False) def main(args: argparse.Namespace):