liaopeiyuan commented on pull request #7719: URL: https://github.com/apache/tvm/pull/7719#issuecomment-855392525
Does the onnxruntime build on CI come with training mode enabled? Seems like the problem is onnxruntime being unhappy with batchnorm containing multiple outputs wuile `training_mode=False` by default. And, taking a step back at this PR, what would be our ideal behavior for batchnorm if the user bring an ONNX model in training mode? Do we also include the extra `mean` and `var` in outputs, or do we throw them away for them? If the latter is what we want, then it doesn't make sense to write a unit test case since ONNX and TVM behavior will be different. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
