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, 11:00 UTC
[Bugfix][CPU] Fix broken AVX2 CPU TP support (#17252)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
May 9, 2025 at 16:55 UTC
6e4a93e3f78a6af2447328d1e7db0fa66f0bc5be
1 file changed
+2
-1
vllm/distributed/device_communicators/cpu_communicator.py
+2
-1
@@ -22,7 +22,8 @@ class CpuCommunicator(DeviceCommunicatorBase):
22
super().__init__(cpu_group, device, device_group, unique_name)
23
self.dist_module = torch.distributed
24
25
- if current_platform.get_cpu_architecture() == CpuArchEnum.X86:
25
+ if (current_platform.get_cpu_architecture() == CpuArchEnum.X86) \
26
+ and hasattr(torch.ops._C, "init_shm_manager"):
27
self.dist_module = _CPUSHMDistributed(self)
28
29
def all_reduce(self, input_):