AttributeError: module 'torch.distributed' has no attribute 'P2POp'

msr-fiddle / pipedream

MIT License

379 stars 117 forks source link

AttributeError: module 'torch.distributed' has no attribute 'P2POp' #75

Closed guanyonglai closed 1 year ago

guanyonglai commented 1 year ago

Why always appear this error, I have according to the readme: docker pull nvcr.io/nvidia/pytorch:20.03-py3, use docker to run pp2bw train, but still raise the wrong

guanyonglai commented 1 year ago

it should use docker 22.12: nvidia-docker pull nvcr.io/nvidia/pytorch:22.12-py3