t-vi opened a new pull request #5756: URL: https://github.com/apache/incubator-tvm/pull/5756
Numpy interprets float as float64. As we drop to numpy for determining what a dtype argument means, this means we inadvertently create float64 consts. This is particularly relevent when trying to create a const matching a dtype of an existing object. This brings creating consts in line with TVM usances. I don't really have an idea where to test this (const creation). I'll be pushing an amendment to the PyTorch frontend shortly which would use this feature and will have tests there. Or I would follow advice on where to put a test just for this. ---------------------------------------------------------------- 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]
