[Bugfix] Fix vllm bench ... on CPU-only head nodes (#25283)

Signed-off-by: Aydin Abiar <aydin@anyscale.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Aydin Abiar <aydin@anyscale.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Aydin Abiar
2025-10-08 09:06:42 -07:00
committed by GitHub
parent c1b06fc182
commit 76afe4edf8
3 changed files with 34 additions and 0 deletions

View File

@@ -8,6 +8,11 @@ to avoid certain eager import breakage."""
from __future__ import annotations
import importlib.metadata
import sys
from vllm.logger import init_logger
logger = init_logger(__name__)
def main():
@@ -29,6 +34,22 @@ def main():
cli_env_setup()
# For 'vllm bench *': use CPU instead of UnspecifiedPlatform by default
if len(sys.argv) > 1 and sys.argv[1] == "bench":
logger.debug(
"Bench command detected, must ensure current platform is not "
"UnspecifiedPlatform to avoid device type inference error"
)
from vllm import platforms
if platforms.current_platform.is_unspecified():
from vllm.platforms.cpu import CpuPlatform
platforms.current_platform = CpuPlatform()
logger.info(
"Unspecified platform detected, switching to CPU Platform instead."
)
parser = FlexibleArgumentParser(
description="vLLM CLI",
epilog=VLLM_SUBCMD_PARSER_EPILOG.format(subcmd="[subcommand]"),