[Hardware] [Intel] Enable Multiprocessing and tensor parallel in CPU backend and update documentation (#6125)
This commit is contained in:
@@ -296,6 +296,9 @@ class GroupCoordinator:
|
||||
pynccl_comm = self.pynccl_comm
|
||||
if (pynccl_comm is not None and not pynccl_comm.disabled):
|
||||
pynccl_comm.all_reduce(input_)
|
||||
elif input_.is_cpu:
|
||||
import intel_extension_for_pytorch as ipex
|
||||
ipex.distributed.all_reduce(input_, group=self.device_group)
|
||||
else:
|
||||
torch.distributed.all_reduce(input_, group=self.device_group)
|
||||
return input_
|
||||
|
||||
Reference in New Issue
Block a user