tqchen opened a new issue #8579: URL: https://github.com/apache/tvm/issues/8579
Looking at the test time log, https://ci.tlcpack.ai/job/tvm/job/main/1356/testReport/ctypes.tests.python.relay/test_op_grad_level2/ conv2d grad takes a long time to finish(1h). This is likely due to our numerical gradient checker trying to do numerical gradient for all directions which scales linearly to the number of entries. Possible resolutions: - K0: Use memoization to cache the numerical gradient needs. - K1: run a different kind of test that randomizes the direction and check consistency, see page 8 of this [slide](https://catalyst.cs.cmu.edu/15-884-mlsys-sp21/slides/2-MLSys-AutomaticDifferentiation.pdf) I would suggest we go with K1 and have an option in check numercial gradient to do so. -- 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]
