guberti opened a new issue, #13364: URL: https://github.com/apache/tvm/issues/13364
When the microTVM team was working on our submission for MLPerf Tiny, we noticed the `kws` model had slightly worse accuracy than expected (we classified `897/1000` samples correctly, while folks are expected to classify [at least `900/1000` correctly](https://github.com/mlcommons/tiny/tree/master/benchmark)). This issue does not occur when running with CMSIS-NN schedules. Despite a bit of looking, it is unclear to me what the issue is. It's possible that it's a padding issue, or maybe a problem with requantization? Total guesses. -- 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]
