Module 'torch.distributed' has no attribute 'ReduceOp'

I try "torch-2.0.0+nv23.05-cp38-cp38-linux_aarch64.whl " and “torch-2.0.0a0+fe05266f.nv23.04-cp38-cp38-linux_aarch64.whl” as AttributeError: module 'torch.distributed' has no attribute 'ReduceOp' ,but still failed .
my versions as below :
jetpack is 5.1.1

$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Sun_Oct_23_22:16:07_PDT_2022
Cuda compilation tools, release 11.4, V11.4.315
Build cuda_11.4.r11.4/compiler.31964100_0

11.4/compiler.31964100_0
qiangsir@ubuntu:~$ jetson_release
‘DISPLAY’ environment variable not set… skipping surface info

  • Jetson AGX Orin
    • Jetpack UNKNOWN [L4T 35.3.1]
    • NV Power Mode: MAXN - Type: 0
    • jetson_stats.service: active
  • Libraries:
    • CUDA: NOT_INSTALLED
    • cuDNN: 8.6.0.166
    • TensorRT: 8.5.2.2
    • Visionworks: NOT_INSTALLED
    • OpenCV: 4.5.4 compiled CUDA: NO
    • VPI: 2.2.7
    • Vulkan: 1.3.204

Hi,

Have you run the script on another platform before?
If yes, which PyTorch version is used?

Thanks.

I want to run renet50 on my jetson agx orin 64G , when I install mmpretrain from source (skip conda) by https://mmpretrain.readthedocs.io/en/latest/index.html, it failed as :

while I flash my jetson agx orin 64 to jetpack 5.1 ,this problem happens again

@qiangqiangsir the PyTorch 1.11 wheel was the last one to be built with USE_DISTRIBUTED:

PyTorch v1.11.0

If you require a newer version of PyTorch with distributed enabled, please see this thread for instructions on building PyTorch from source:

Or perhaps it’s possible to disable distributed mode in the mmpretrain library you are using?

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.