Error occurs in rpc-pipeline code in tutorial

I am using torch 1.6. When I try to run the code, it tells me like below:

AttributeError: ‘torch.distributed.rpc.TensorPipeRpcBackendOptions’ object has no attribute ‘num_send_recv_threads’

Could anyone please tell me what possible reasons could be?

details are like below:
File “rpc_pipeline.py”, line 291, in setup
rpc_backend_options=options
File “/conda/lib/python3.7/site-packages/torch/distributed/rpc/init.py", line 90, in init_rpc
rpc_backend_options=options
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/init.py”, line 90, in init_rpc
api._init_rpc_backend(backend, store, name, rank, world_size, rpc_backend_options)
File “/conda/lib/python3.7/site-packages/torch/distributed/rpc/api.py", line 299, in _init_rpc_backend
api._init_rpc_backend(backend, store, name, rank, world_size, rpc_backend_options)
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/api.py”, line 299, in _init_rpc_backend
api._init_rpc_backend(backend, store, name, rank, world_size, rpc_backend_options)
File “/conda/lib/python3.7/site-packages/torch/distributed/rpc/api.py", line 299, in _init_rpc_backend
rpc_backend_options=rpc_backend_options,
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py”, line 94, in init_backend
rpc_backend_options=rpc_backend_options,
File “/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py", line 94, in init_backend
rpc_backend_options=rpc_backend_options,
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py”, line 94, in init_backend
return backend.value.init_backend_handler(args, **kwargs)
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py", line 144, in _process_group_init_backend_handler
return backend.value.init_backend_handler(args, **kwargs)
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py", line 144, in _process_group_init_backend_handler
rpc_backend_options.num_send_recv_threads,
AttributeError: ‘torch.distributed.rpc.TensorPipeRpcBackendOptions’ object has no attribute ‘num_send_recv_threads’
rpc_backend_options.num_send_recv_threads,
AttributeError: ‘torch.distributed.rpc.TensorPipeRpcBackendOptions’ object has no attribute ‘num_send_recv_threads’
return backend.value.init_backend_handler(args, **kwargs)
File "
/conda/lib/python3.7/site-packages/torch/distributed/rpc/backend_registry.py", line 144, in _process_group_init_backend_handler
rpc_backend_options.num_send_recv_threads,

I solve this after adding this argument.
image

1 Like

Yep, we are switching to TensorPipe as the default RPC backend in v1.7.