def test_bert_model(self, version):
model_spec = ms.BertQAModelSpec(trainable=False, predict_batch_size=1)
train_data, validation_data = _get_data(model_spec, version)
model = question_answer.create(train_data,
model_spec=model_spec,
epochs=1,
batch_size=1)
self._test_f1_score(model, validation_data, 0.0)
self._test_export_vocab(model)
self._test_export_to_tflite(model, validation_data)
self._test_export_to_saved_model(model)
def test_bert_model(self):
# Only test squad1.1 since it takes too long time for this.
version = '1.1'
model_spec = ms.BertQAModelSpec(trainable=False, predict_batch_size=1)
train_data, validation_data = _get_data(model_spec, version)
model = question_answer.create(
train_data, model_spec=model_spec, epochs=1, batch_size=1)
self._test_f1_score(model, validation_data, 0.0)
self._test_export_vocab(model)
self._test_export_to_tflite(
model, validation_data, expected_json_file='bert_qa_metadata.json')
self._test_export_to_saved_model(model)
def test_bert_model_v1_incompatible(self):
with self.assertRaisesRegex(ValueError, 'Incompatible versions'):
_ = ms.BertQAModelSpec(trainable=False)
