Conversation
`py::array::itemsize()` returns 0 for tests in `model_test.py`. I could not really figure out why from pybind11 docs. Using `request().itemsize` on the other hand always returns the correct item size.
Adds CMakeLists.txt file for cmake build of onnx app. The build logic is kept same as the one from corresponding Makefile.
|
Thanks for taking a crack at this! We're currently performing some maintenance on the build bots, but I will take a look at this when we're done. |
|
Going to see what the buildbots think. Looking at recent logs, it seems There probably needs to be a |
alexreinking
left a comment
There was a problem hiding this comment.
Still waiting on buildbots, but this is what stood out to me on a first pass.
|
Some of the buildbots were missing |
|
Thanks for the review and the follow up fixes. I will look into addressing the comments soon :)
Ah okay. Should I skip building this app if |
|
@vawale -- while playing with this, I discovered a bunch more changes that needed to be made in other parts of the project. Assuming the buildbots like it at this point, I'm happy to merge. |
|
New failures on the mac bots because WebGPU is missing. This is because they're running under launchd now and don't have the environment variables set. This is brittle, so I'm going to do this The Right Way... |
Turns out this is too hard. See #8714 . |
|
@vawale -- I see that the halide_as_onnx_backend_test is timing out (and that subtests are failing). Does it work for you? https://buildbot.halide-lang.org/master/#/builders/265/builds/30 |
|
Seems like |
Yes, I observe the same. The auto-scheduler takes too long. |
|
Despite all the workflows failing... none of them are related to this change. Thanks so much for your help! |
Adds CMakeLists.txt file for cmake build of onnx app. The build logic is kept same as the one from corresponding Makefile.
Fixes #8737