def _predict_without_distribute_strategy(self, model, input_fn):
"""Predicts the dataset without using distribute strategy."""
ds = input_fn()
all_results = []
for features, _ in ds:
outputs = model.predict_on_batch(features)
for unique_id, start_logits, end_logits in zip(features['unique_ids'],
outputs[0], outputs[1]):
raw_result = run_squad_helper.RawResult(
unique_id=unique_id.numpy(),
start_logits=start_logits.tolist(),
end_logits=end_logits.tolist())
all_results.append(raw_result)
if len(all_results) % 100 == 0:
tf.compat.v1.logging.info('Made predictions for %d records.',
len(all_results))
return all_results
def predict_tflite(self, tflite_filepath, dataset):
"""Predicts the dataset for TFLite model in `tflite_filepath`."""
all_results = []
lite_runner = model_util.LiteRunner(tflite_filepath,
self.reorder_input_details,
self.reorder_output_details)
for features, _ in dataset:
outputs = lite_runner.run(features)
for unique_id, start_logits, end_logits in zip(
features['unique_ids'], outputs[0], outputs[1]):
raw_result = run_squad_helper.RawResult(
unique_id=unique_id.numpy(),
start_logits=start_logits.tolist(),
end_logits=end_logits.tolist())
all_results.append(raw_result)
if len(all_results) % 100 == 0:
tf.compat.v1.logging.info(
'Made predictions for %d records.', len(all_results))
return all_results
