TensorFlow Lite model coverage tests
Various conversion tests on popular mobile models.
Golden values
Some tests rely on pre-computed golden values. The main goal is to detect changes affecting unintended parts of TFLite.
Should a golden value test fail after an intended change, the golden values can be updated with the following command:
bazel run //third_party/tensorflow/lite/testing/model_coverage:<target> --test_output=all -- --update_goldens
Notice bazel run instead of bazel test and the addition of the
--update_golden flag.
The updated golden data files must then be included in the change list.