masahi commented on pull request #7513: URL: https://github.com/apache/tvm/pull/7513#issuecomment-785492581
The point is, no matter how we try to support `copy_`, there seems to be an edge case pytorch usage that prevents a correct translation to Relay. See the discussion and examples in the PR linked above. Rather than adding `copy_` support that sometime works and sometime doesn't, the current consensus seems to be that we shouldn't support this op. The best way forward is to remove that inplace assignment idiom from your models. Don't say you cannot change the model because of customer issues, that's not our problem. ---------------------------------------------------------------- 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]
