merrymercy edited a comment on pull request #6671: URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-716905146
The auto-scheduler part looks good to me. I believe this PR solves the problem while not bringing any performance regression. But I am not sure about the autotvm part. Your test case is not correct. With `n-trials=16`, the code modified by you will not even be executed. Because the cost model only starts to work after getting enough training samples (n-trials > 64). To do the benchmark correctly, you should use a larger matrix size (n=1024), run the autotvm with n-trials > 64 and `XGBTuner`, and report the time used in simulated annealing (i.e., the part modified by you). Could you delete the autotvm part in this PR so we can merge 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: us...@infra.apache.org
