masahi commented on issue #8679: URL: https://github.com/apache/tvm/issues/8679#issuecomment-895668415
I can confirm that `model.pt` and a TVM model converted via ONNX give the same output. It is hard to compare two TVM models, one coming from PT frontend and the other from ONNX, since the ONNX model folds batch norm into convolution so there is no batch norm in ONNX model. Since the differences from resnet18 seem small, and we know that there is no issue with resnet18, maybe you can start with resnet18 and gradually add your changes until the TVM result becomes off? I'm using a script modified from https://github.com/apache/tvm/blob/main/tutorials/frontend/from_pytorch.py to test the accuracy, and there is no need to train the model. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
