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 3, 08:00 UTC
[XPU] Use custom op collective behavior (#41354)
Signed-off-by: Chaojun,Zhang <chaojun.zhang@intel.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
Chaojun Zhang committed
May 19, 2026 at 14:14 UTC
f1e3f0e6d685082bdb313c20914099ac5ede5f14
1 file changed
+4
vllm/platforms/xpu.py
+4
@@ -409,3 +409,7 @@ class XPUPlatform(Platform):
409
@classmethod
410
def num_compute_units(cls, device_id: int = 0) -> int:
411
return torch.xpu.get_device_properties(device_id).max_compute_units
412
+
413
+ @classmethod
414
+ def use_custom_op_collectives(cls) -> bool:
415
+ return True