Skip to content

Commit 82ccece

Browse files
authored
Install the yolo26 requirements in the Samsung models CI job (pytorch#23452)
The `test-samsung-models-linux` job on main has failed on every run since pytorch#22960 added `test_yolo26.py`. The test imports the yolo26 example wrapper, and the wrapper imports `ultralytics`. The job never installs it, so that test module fails to import: ``` ModuleNotFoundError: No module named 'ultralytics' ``` `unittest` reports the broken module as one error and still runs the other tests, but the error is enough to turn the whole job red. That import error is in every failing log I checked. Some runs also have device connection errors (`edb command execute failed`) in other tests. This change installs `examples/models/yolo26/requirements.txt` in that job before the tests run. It uses the same flags and requirements file as `test_yolo26.sh` and the yolo26 README, which keep the torch that CI already has. pytorch#23449 adds the same step for the generic yolo26 portable job. On pull requests, this job only runs when the PR changes the PyTorch pin, `.ci/scripts`, or `zephyr`, or when it has the `ciflow/trunk` label. pytorch#22960 changed none of those, which is likely why it merged without this job running. This PR runs it through the label. ### This does not make the job green yet With the package installed, `test_yolo26` now runs, and it fails inside the Samsung compiler (AI LiteCore v1.2.0): ``` [Exynos AI LiteCore][ERROR] cannot find exporter for layer type : CUCustomOp [ERROR][Exynos ExecuTorch][PyEnnWrapperAdaptor.h:66]: Compile model failed. AssertionError (enn_preprocess.py, assert enn_context_binary is not None ...) ``` The missing import was hiding this failure, so CI has never run this test to the end. The other 12 Samsung model tests pass in this run. This needs a fix on the Samsung side, either in the yolo26 lowering or in the SDK, tracked in pytorch#23481. pytorch#23351 moves CI to SDK v1.2.1. I have not checked whether that version can export this layer. ### Test plan - The edited workflow file parses, and the job script passes `bash -n`. - In the Samsung job on this PR, `ultralytics` 8.4.14 installed and torch stayed the same. Then `test_yolo26` failed with the compiler error above: `Ran 13 tests`, `FAILED (failures=1)`.
1 parent 9a9f422 commit 82ccece

1 file changed

Lines changed: 3 additions & 0 deletions

File tree

‎.github/workflows/trunk.yml‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1147,4 +1147,7 @@ jobs:
11471147
# accelerate is for test_mobilebert_qat, whose TrainingArguments needs it;
11481148
# transformers reports the same error when it is absent and when it is too old.
11491149
pip install evaluate 'accelerate>=1.1.0'
1150+
pip install --upgrade-strategy only-if-needed \
1151+
--extra-index-url https://download.pytorch.org/whl/cpu \
1152+
-r examples/models/yolo26/requirements.txt
11501153
python -m unittest discover -s backends/samsung/test/models -p "test_*.py"

0 commit comments

Comments
 (0)