DeepSpeed

[CPU] Allow deepspeed.comm.inference_all_reduce in torch.compile graph

#5604

Merged