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 5, 02:15 UTC
[XPU] use spawn multiproc method on xpu (#39671)
Signed-off-by: Xinyu Chen <xinyu1.chen@intel.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
Xinyu Chen committed
Apr 16, 2026 at 14:42 UTC
f4ddaf8cf7f7daf1233acbfd4db97ca107a0a5b7
1 file changed
+4
vllm/platforms/xpu.py
+4
@@ -228,6 +228,10 @@ class XPUPlatform(Platform):
228
# ref. https://openucx.readthedocs.io/en/master/faq.html
229
os.environ["UCX_MEMTYPE_CACHE"] = "n"
230
231
+ # spawn is the only supported multiprocessing method on XPU
232
+ if "VLLM_WORKER_MULTIPROC_METHOD" not in os.environ:
233
+ os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
234
+
235
@classmethod
236
def update_block_size_for_backend(cls, vllm_config: "VllmConfig") -> None:
237
super().update_block_size_for_backend(vllm_config)