new shape sometime is input tensor in TFLite model file.
That's why the tflite parser will fail randomly since the op options will be None in this case.
Manually implementing this hot fix then running the unit tests in #133 shows this fix corrects the breaking test.
See test_api/test_keras_tflite/test_export.py test_keras_model
new shape sometime is input tensor in TFLite model file. That's why the tflite parser will fail randomly since the op options will be
None
in this case.