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, 06:00 UTC
[Misc] Remove duplicate multiproc method setting for CPU platform (#19649)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Jun 16, 2025 at 10:26 UTC
a5e7242d5f4e88182e6332696c0113e7ce8467a0
1 file changed
-3
vllm/platforms/cpu.py
-3
@@ -216,9 +216,6 @@ class CpuPlatform(Platform):
216
# Disable torch async compiling which won't work with daemonic processes
217
os.environ["TORCHINDUCTOR_COMPILE_THREADS"] = "1"
218
219
- # Share the cpusets list among ranks by spawning process instead
220
- os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
221
-
219
# Intel OpenMP setting
220
ld_prealod_str = os.getenv("LD_PRELOAD", "")
221
if "libiomp5.so" in ld_prealod_str: