tflite_model_maker.config.QuantizationConfig

Configuration for post-training quantization.

tflite_model_maker.config.QuantizationConfig(
    optimizations=None,
    representative_data=None,
    quantization_steps=None,
    inference_input_type=None,
    inference_output_type=None,
    supported_ops=None,
    supported_types=None,
    experimental_new_quantizer=None
)

Used in the notebooks

Used in the tutorials

Refer to https://www.tensorflow.org/lite/performance/post_training_quantization for different post-training quantization options.

optimizations A list of optimizations to apply when converting the model. If not set, use [Optimize.DEFAULT] by default.
representative_data A DataLoader holding representative data for post-training quantization.
quantization_steps Number of post-training quantization calibration steps to run.
inference_input_type Target data type of real-number input arrays. Allows for a different type for input arrays. Defaults to None. If set, must be be {tf.float32, tf.uint8, tf.int8}.
inference_output_type Target data type of real-number output arrays. Allows for a different type for output arrays. Defaults to None. If set, must be {tf.float32, tf.uint8, tf.int8}.
supported_ops Set of OpsSet options supported by the device. Used to Set converter.target_spec.supported_ops.
supported_types List of types for constant values on the target device. Supported values are types exported by lite.constants. Frequently, an optimization choice is driven by the most compact (i.e. smallest) type in this list (default [constants.FLOAT]).
experimental_new_quantizer Whether to enable experimental new quantizer.

Methods

for_dynamic

View source

@classmethod
for_dynamic()

Creates configuration for dynamic range quantization.

for_float16

View source

@classmethod
for_float16()

Creates configuration for float16 quantization.

for_int8

View source

@classmethod
for_int8(
    representative_data,
    quantization_steps=DEFAULT_QUANTIZATION_STEPS,
    inference_input_type=tf.uint8,
    inference_output_type=tf.uint8,
    supported_ops=tf.lite.OpsSet.TFLITE_BUILTINS_INT8
)

Creates configuration for full integer quantization.

Args
representative_data Representative data used for post-training quantization.
quantization_steps Number of post-training quantization calibration steps to run.
inference_input_type Target data type of real-number input arrays. Used only when is_integer_only is True. Must be in {tf.uint8, tf.int8}.
inference_output_type Target data type of real-number output arrays. Used only when is_integer_only is True. Must be in {tf.uint8, tf.int8}.
supported_ops Set of tf.lite.OpsSet options, where each option represents a set of operators supported by the target device.

Returns
QuantizationConfig.

get_converter_with_quantization

View source

get_converter_with_quantization(
    converter, **kwargs
)

Gets TFLite converter with settings for quantization.