torch.distributed

API名称

是否支持

限制与说明

torch.distributed.destroy_process_group

  
  • 不推荐使用该接口,当前显示使用该接口销毁processgroup对象时可能会出现进程卡住的问题,可参考PyTorch原生issues信息https://github.com/pytorch/pytorch/issues/75097
  • 如使用该接口,请确保torch.npu.set_device在torch.distributed.init_process_group之前调用,否则可能会出现进程卡住的问题。

torch.distributed.is_available

  

torch.distributed.init_process_group

  

torch.distributed.is_initialized

  

torch.distributed.is_mpi_available

  

torch.distributed.is_nccl_available

  

torch.distributed.is_torchelastic_launched

  

torch.distributed.Backend

  

torch.distributed.get_backend

  

torch.distributed.get_rank

  

torch.distributed.get_world_size

  

torch.distributed.Store

  

torch.distributed.TCPStore

  

torch.distributed.HashStore

  

torch.distributed.FileStore

  

torch.distributed.PrefixStore

  

torch.distributed.Store.set

  

torch.distributed.Store.get

  

torch.distributed.Store.add

  

torch.distributed.Store.compare_set

  

torch.distributed.Store.wait

  

torch.distributed.Store.num_keys

  

torch.distributed.Store.delete_key

  

torch.distributed.Store.set_timeout

  

torch.distributed.new_group

  

torch.distributed.send

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.distributed.recv

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.distributed.isend

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.distributed.irecv

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.distributed.broadcast

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.distributed.broadcast_object_list

  

torch.distributed.all_reduce

支持fp16, fp32, int32, int64, bool

torch.distributed.reduce

支持bf16,fp16,fp32,uint8,int8,int32,int64,bool

torch.distributed.all_gather

支持bf16,fp16,fp32,int8,int32,bool

torch.distributed.all_gather_object

  

torch.distributed.gather

  

torch.distributed.gather_object

  

torch.distributed.scatter

  

torch.distributed.scatter_object_list

  

torch.distributed.reduce_scatter

支持bf16,fp16,fp32,int8,int32

torch.distributed.all_to_all

支持fp32

torch.distributed.barrier

  

torch.distributed.monitored_barrier

  

torch.distributed.ReduceOp

支持bf16,fp16,fp32,uint8,int8,int32,int64,bool

torch.distributed.reduce_op

支持bf16,fp16,fp32,uint8,int8,int32,int64,bool