manupa-arm commented on issue #10487: URL: https://github.com/apache/tvm/issues/10487#issuecomment-1063105756
@driazati, > Sure we can do that moving forward, but if a test fails on more than 1 parameterized version I think we should still stick with disabling the whole test to reduce the chance of further flaky failures. I dont have a strong view eitherway, dependy on the policy going forward, we might restructure the tests. > No particular reason other than the tests are in a similar category, but we can be more granular with reporting in the future (so 1-issue-per-test pretty much, unless errors are the same) I think its better to report errors that are similiar because the work done to correct these are orthogonal. > https://ci.tlcpack.ai/blue/organizations/jenkins/tvm/detail/main/2671/pipeline/ This one seems like corrupt download. TVM has a lot of tests that downloads models for tests and they can fail due to external reasons (i.e. network, CI node limitations). Either we need to decide against using such tests (i.e. holding binary models in-tree) or be a bit lenient on network failures (accepting that they happen once in a while), because I dont see a way to resolve this issue. -- 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]
