XFAIL failing tflite test

[nirvedhmeshram]

The datatset for some tests doesnt seem currently accessible. More details on
#18518

[ScottTodd]

Do you want to disable iree/tests/e2e/stablehlo_models/mnist_train_test/mnist_train_test_cpu in this PR too?

Logs: https://github.com/iree-org/iree/actions/runs/11805834745/job/32889192224#step:7:2389

URL (same location that is no longer accessible):

iree/tests/e2e/stablehlo_models/mnist_train_test/mnist_train_test.py

Line 22 in e0cd5b8

MODEL_ARTIFACTS_URL = "https://storage.googleapis.com/iree-model-artifacts/mnist_train.2bec0cb356ae7c059e04624a627eb3b15b0a556cbd781bbed9f8d32e80a4311d.tar"

[ScottTodd]

While XFAIL is generally useful, I would actually exclude the test here with REQUIRES, using this pattern:

iree/integrations/tensorflow/test/iree_tfl_tests/cartoon_gan.run

Lines 1 to 3 in e0cd5b8

	# REQUIRES: bugfix
	# tfhub/kaggle broke their download endpoint, update URLs
	# RUN: %PYTHON -m iree_tfl_tests.cartoon_gan_test --artifacts_dir=%t

The specific file this test is trying to download is

iree/integrations/tensorflow/test/python/iree_tfl_tests/mobilenet_v1_test.py

Line 11 in e0cd5b8

model_path = "https://storage.googleapis.com/iree-model-artifacts/tflite-integration-tests/mobilenet_v1.tflite"

That URL is returning an error. We have a tracking issue covering migrating iree-model-artifacts GCP files to other locations: #18518 .

I'd rather keep the test disabled until we take intentional action to move the file, instead of XFAIL it and then have the test maybe start unexpectedly passing if the URL becomes live again.

[ScottTodd]

#19127 will supersede this (though I didn't run the integrations/tensorflow/test/iree_tfl_tests/update_tflite_model_documentation.py script)

[nirvedhmeshram]

#19127 will supersede this (though I didn't run the integrations/tensorflow/test/iree_tfl_tests/update_tflite_model_documentation.py script)

Sounds good. I can keep that part and refactor this PR tomorrow.

[ScottTodd]

#19127 will supersede this (though I didn't run the integrations/tensorflow/test/iree_tfl_tests/update_tflite_model_documentation.py script)

Sounds good. I can keep that part and refactor this PR tomorrow.

That's fine, thanks. A better fix than just running the script as-is would be updating the tests to use files that are still available, so the test status is accurate (we should support mobilenet_v1 generally, even if our mirror of the .tflite file for mobilenet_v1 used in a test is inaccessible ;P). I have iree-org/iree-test-suites#5 for the long term solution for such tests.

[nirvedhmeshram]

After speaking to Scott offline, I think the way the state of these tests is currently is what it should be until the long term fixes start landing so I am not going to make any changes.