mbrookhart commented on pull request #7550: URL: https://github.com/apache/tvm/pull/7550#issuecomment-787634639
I thought about this over the weekend. I agree that this is the right solution. Fold-Constant doesn't assume training or inference, it's more general, and in the training case, dropout should never be folded. Additionally, the VM and the GraphRuntime both run FoldConstant after SimplifyInference, so this shouldn't hurt inference performance. ---------------------------------------------------------------- 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]
