How to build and run "ModelTests" unit tests #7660
-
|
I have one of my pull requests that is failing due to the "ModelTests" unit tests in the CI build. The failure was really clear and easy to resolve. However, I have not been able to figure out how to run the "ModelTests" as part of a local build. I have found the code located in If I had seen the failures locally I would not have wasted CI build time. I would like to know how to build and run these tests locally so I can run them against future commits. Is there a build flag that I am unaware of that will cause the "ModelTests" to build when onnxruntime_test_all is built? Is this set of tests only limited to Linux OS or can I also build and run them on Windows? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
To run it locally, you need to modify the .cc a little bit and enable "--enable_onnx_tests" in the build command. However, since the input data is not in public, so you won't be able to run it locally. In fact, the model input is proprietary and private, so we won't be able to share it. |
Beta Was this translation helpful? Give feedback.
-
That is unfortunate but understandable. Thanks for the reply @ytaous. I guess I will have to rely on the CI build system. At least the error was clear and I was able to resolve the issue without having access to the tests. :) |
Beta Was this translation helpful? Give feedback.
To run it locally, you need to modify the .cc a little bit and enable "--enable_onnx_tests" in the build command. However, since the input data is not in public, so you won't be able to run it locally.
Example:
./build.sh --config Debug --use_cuda --enable_training --build_wheel --skip_tests --parallel 8 --enable_onnx_tests
./onnxruntime_test_all --gtest_filter=ModelTests*
In fact, the model input is proprietary and private, so we won't be able to share it.