self.uri = uri
self.compat_tf_versions = compat.get_compat_tf_versions(
compat_tf_versions)
self.name = name
if input_image_shape is None:
input_image_shape = [224, 224]
self.input_image_shape = input_image_shape
mobilenet_v2_spec = functools.partial(
ImageModelSpec,
uri='https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/4',
compat_tf_versions=2,
name='mobilenet_v2')
mobilenet_v2_spec.__doc__ = util.wrap_doc(ImageModelSpec,
'Creates MobileNet v2 model spec.')
mm_export('image_classifier.MobileNetV2Spec').export_constant(
__name__, 'mobilenet_v2_spec')
resnet_50_spec = functools.partial(
ImageModelSpec,
uri='https://tfhub.dev/google/imagenet/resnet_v2_50/feature_vector/4',
compat_tf_versions=2,
name='resnet_50')
resnet_50_spec.__doc__ = util.wrap_doc(ImageModelSpec,
'Creates ResNet 50 model spec.')
mm_export('image_classifier.Resnet50Spec').export_constant(
__name__, 'resnet_50_spec')
efficientnet_lite0_spec = functools.partial(
ImageModelSpec,
eval_metrics = squad_evaluate_v1_1.evaluate(
pred_dataset, all_predictions)
return eval_metrics
mobilebert_classifier_spec = functools.partial(
BertClassifierModelSpec,
uri=
'https://tfhub.dev/google/mobilebert/uncased_L-24_H-128_B-512_A-4_F-4_OPT/1',
is_tf2=False,
distribution_strategy='off',
name='MobileBert',
default_batch_size=48,
)
mobilebert_classifier_spec.__doc__ = util.wrap_doc(
BertClassifierModelSpec,
'Creates MobileBert model spec for the text classification task. See also: `tflite_model_maker.text_classifier.BertClassifierSpec`.'
)
mm_export('text_classifier.MobileBertClassifierSpec').export_constant(
__name__, 'mobilebert_classifier_spec')
mobilebert_qa_spec = functools.partial(
BertQAModelSpec,
uri=
'https://tfhub.dev/google/mobilebert/uncased_L-24_H-128_B-512_A-4_F-4_OPT/1',
is_tf2=False,
distribution_strategy='off',
learning_rate=4e-05,
name='MobileBert',
default_batch_size=32,
)
mobilebert_qa_spec.__doc__ = util.wrap_doc(
tf.lite.OpsSet.TFLITE_BUILTINS
]
tflite_model = converter.convert()
with tf.io.gfile.GFile(tflite_filepath, 'wb') as f:
f.write(tflite_model)
efficientdet_lite0_spec = functools.partial(
EfficientDetModelSpec,
model_name='efficientdet-lite0',
uri='https://tfhub.dev/tensorflow/efficientdet/lite0/feature-vector/1',
)
efficientdet_lite0_spec.__doc__ = util.wrap_doc(
EfficientDetModelSpec,
'Creates EfficientDet-Lite0 model spec. See also: `tflite_model_maker.object_detector.EfficientDetSpec`.'
)
mm_export('object_detector.EfficientDetLite0Spec').export_constant(
__name__, 'efficientdet_lite0_spec')
efficientdet_lite1_spec = functools.partial(
EfficientDetModelSpec,
model_name='efficientdet-lite1',
uri='https://tfhub.dev/tensorflow/efficientdet/lite1/feature-vector/1',
)
efficientdet_lite1_spec.__doc__ = util.wrap_doc(
EfficientDetModelSpec,
'Creates EfficientDet-Lite1 model spec. See also: `tflite_model_maker.object_detector.EfficientDetSpec`.'
)
mm_export('object_detector.EfficientDetLite1Spec').export_constant(
__name__, 'efficientdet_lite1_spec')
def bert_qa_spec(**kwargs):
return BertQAModelSpec(**kwargs)
mobilebert_classifier_spec = functools.partial(
BertClassifierModelSpec,
uri=
'https://tfhub.dev/google/mobilebert/uncased_L-24_H-128_B-512_A-4_F-4_OPT/1',
is_tf2=False,
distribution_strategy='off',
name='MobileBert',
default_batch_size=48,
)
mobilebert_classifier_spec.__doc__ = util.wrap_doc(
BertClassifierModelSpec,
'Creates MobileBert model spec for the text classification task.')
mm_export('text_classifier.MobileBertClassifierSpec').export_constant(
__name__, 'mobilebert_classifier_spec')
mobilebert_qa_spec = functools.partial(
BertQAModelSpec,
uri=
'https://tfhub.dev/google/mobilebert/uncased_L-24_H-128_B-512_A-4_F-4_OPT/1',
is_tf2=False,
distribution_strategy='off',
learning_rate=4e-05,
name='MobileBert',
default_batch_size=32,
)
mobilebert_qa_spec.__doc__ = util.wrap_doc(
'eval_top_k': eval_top_k,
}
def create_model(self):
"""Creates recommendation model based on params.
Returns:
Keras model.
"""
return _rm.RecommendationModel(self.params)
recommendation_bow_spec = functools.partial(RecommendationSpec,
encoder_type='bow')
recommendation_bow_spec.__doc__ = util.wrap_doc(
RecommendationSpec,
'Creates Recommendation Bag-of-Word (BoW) model spec. See also: `tflite_model_maker.recommendation.ModelSpec`.'
)
mm_export('recommendation.BowSpec').export_constant(__name__,
'recommendation_bow_spec')
recommendation_cnn_spec = functools.partial(RecommendationSpec,
encoder_type='cnn')
recommendation_cnn_spec.__doc__ = util.wrap_doc(
RecommendationSpec,
'Creates Recommendation CNN model spec. See also: `tflite_model_maker.recommendation.ModelSpec`.'
)
mm_export('recommendation.CnnSpec').export_constant(__name__,
'recommendation_cnn_spec')
recommendation_rnn_spec = functools.partial(RecommendationSpec,
encoder_type='rnn')
input_image_shape = [224, 224]
self.input_image_shape = input_image_shape
def get_default_quantization_config(self, representative_data):
"""Gets the default quantization configuration."""
config = configs.QuantizationConfig.for_int8(representative_data)
return config
mobilenet_v2_spec = functools.partial(
ImageModelSpec,
uri='https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/4',
compat_tf_versions=2,
name='mobilenet_v2')
mobilenet_v2_spec.__doc__ = util.wrap_doc(
ImageModelSpec,
'Creates MobileNet v2 model spec. See also: `tflite_model_maker.image_classifier.ModelSpec`.'
)
mm_export('image_classifier.MobileNetV2Spec').export_constant(
__name__, 'mobilenet_v2_spec')
resnet_50_spec = functools.partial(
ImageModelSpec,
uri='https://tfhub.dev/google/imagenet/resnet_v2_50/feature_vector/4',
compat_tf_versions=2,
name='resnet_50')
resnet_50_spec.__doc__ = util.wrap_doc(
ImageModelSpec,
'Creates ResNet 50 model spec. See also: `tflite_model_maker.image_classifier.ModelSpec`.'
)
mm_export('image_classifier.Resnet50Spec').export_constant(
__name__, 'resnet_50_spec')
'lstm_num_units': lstm_num_units,
'eval_top_k': eval_top_k,
}
def create_model(self):
"""Creates recommendation model based on params.
Returns:
Keras model.
"""
return _rm.RecommendationModel(self.params)
recommendation_bow_spec = functools.partial(
RecommendationSpec, encoder_type='bow')
recommendation_bow_spec.__doc__ = util.wrap_doc(
RecommendationSpec, 'Creates Recommendation Bag-of-Word (BoW) model spec.')
mm_export('recommendation.BowSpec').export_constant(__name__,
'recommendation_bow_spec')
recommendation_cnn_spec = functools.partial(
RecommendationSpec, encoder_type='cnn')
recommendation_cnn_spec.__doc__ = util.wrap_doc(
RecommendationSpec, 'Creates Recommendation CNN model spec.')
mm_export('recommendation.CnnSpec').export_constant(__name__,
'recommendation_cnn_spec')
recommendation_rnn_spec = functools.partial(
RecommendationSpec, encoder_type='rnn')
recommendation_rnn_spec.__doc__ = util.wrap_doc(
RecommendationSpec, 'Creates Recommendation RNN model spec.')
mm_export('recommendation.RnnSpec').export_constant(__name__,
converter.target_spec.supported_ops += [
tf.lite.OpsSet.TFLITE_BUILTINS
]
tflite_model = converter.convert()
with tf.io.gfile.GFile(tflite_filepath, 'wb') as f:
f.write(tflite_model)
efficientdet_lite0_spec = functools.partial(
EfficientDetModelSpec,
model_name='efficientdet-lite0',
uri='https://tfhub.dev/tensorflow/efficientdet/lite0/feature-vector/1',
)
efficientdet_lite0_spec.__doc__ = util.wrap_doc(
EfficientDetModelSpec, 'Creates EfficientDet-Lite0 model spec.')
mm_export('object_detector.EfficientDetLite0Spec').export_constant(
__name__, 'efficientdet_lite0_spec')
efficientdet_lite1_spec = functools.partial(
EfficientDetModelSpec,
model_name='efficientdet-lite1',
uri='https://tfhub.dev/tensorflow/efficientdet/lite1/feature-vector/1',
)
efficientdet_lite1_spec.__doc__ = util.wrap_doc(
EfficientDetModelSpec, 'Creates EfficientDet-Lite1 model spec.')
mm_export('object_detector.EfficientDetLite1Spec').export_constant(
__name__, 'efficientdet_lite1_spec')
efficientdet_lite2_spec = functools.partial(
EfficientDetModelSpec,
