tqchen commented on issue #15204: URL: https://github.com/apache/tvm/issues/15204#issuecomment-1616655567
Thank you for reporting the issue, we are moving towards TensorIR and meta-schedule. and it is expected that some of the legacy te and auto-scheduler may not cover all cases and require tuning to guard against possible failure case. You are more than welcomed to try out TensorIR and meta-schedule instead. For us to work on a fix the test case needs to be actionable and ideally work as a unit test. Meta-schedule also comes with a more friendly trace, with TVMScript so makes it easier for the issue to be actionable. Specifically, most of the test-cases come with a TVMScript before, a trace, and expected tvmscript to know why. Closing this for now as the issue is on a legacy module that do not necessarily guarantee arbitrary schedule robustness, and it is not actionable atm. -- 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]
