dsherry
commented
3 years ago I've been noticing this too lately.
nlp_primitives/tests/test_universal_sentence_encoder.py::test_regular PASSED [ 87%]
nlp_primitives/tests/test_universal_sentence_encoder.py::test_without_tensorflow PASSED [ 88%]
nlp_primitives/tests/test_universal_sentence_encoder.py::test_primitive_serialization PASSED [ 90%]
nlp_primitives/tests/test_universal_sentence_encoder.py::test_feature_serialization make: *** [Makefile:29: unit-tests] Killed
The testing for the universal sentence encoder takes up too much memory, so fails due to being 'killed' on circleci. So, since it passes locally in the same virtualenv setup as circleci, it has been commented out at present. Furthermore, either of the two universal sentence encoder tests can pass in circle ci, just not both at the same time, which further points to the fact that it is a memory usage issue rather than a correctness one.