lanpa/tensorboardX

Test fail.

odidev opened this issue · 3 comments

odidev commented

I have been working to build and test the tensorboardX package for amd64 and arm64 architectures. Followed GitHub Action, while testing this package it is showing the below error for both architectures.

ImportError: cannot import name 'metanet_pb2' from partially initialized module 'caffe2.proto'

Also, I got the same issue which is closed but even after trying provided suggestions still getting the same error.

GHA Log for your reference: https://github.com/odidev/tensorboardX/actions/runs/2165858491

Output at x86_64: tensorboardX_x86_64_result.txt
Output at arm64: tensorboardX_arm64_result.txt

Could you please provide your feedback on it.

odidev commented

@lanpa Is there any update for the above issue ?

lanpa commented

Hi @odidev. As your link indicated, the caffe2 framework might not be tested throughly. I have did a quick test with simply
pip-installed torch with python 3.7, and with python 3.8. Using a simple import statement: from caffe2.proto import caffe2_pb2, metanet_pb2, torch_pb2, the exact error can be reproduced in the 3.8 env while 3.7 is ok. I think it's fine to ignore the caffe2 tests as per pytorch/pytorch#21726 .