Home | History | Annotate | Download | only in optimize 
      1 # Copyright 2019 The TensorFlow Authors. All Rights Reserved.
      2 #
      3 # Licensed under the Apache License, Version 2.0 (the "License");
      4 # you may not use this file except in compliance with the License.
      5 # You may obtain a copy of the License at
      6 #
      7 #     http://www.apache.org/licenses/LICENSE-2.0
      8 #
      9 # Unless required by applicable law or agreed to in writing, software
     10 # distributed under the License is distributed on an "AS IS" BASIS,
     11 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     12 # See the License for the specific language governing permissions and
     13 # limitations under the License.
     14 # ==============================================================================
     15 """Python wrapper for post training quantization with calibration."""
     16 from __future__ import absolute_import
     17 from __future__ import division
     18 from __future__ import print_function
     19 
     20 from tensorflow.python.util.lazy_loader import LazyLoader
     21 
     22 # Lazy load since some of the performance benchmark skylark rules
     23 # break dependencies. Must use double quotes to match code internal rewrite
     24 # rule.
     25 _calibration_wrapper = LazyLoader(
     26     "_calibration_wrapper", globals(),
     27     "tensorflow.lite.python.optimize."
     28     "tensorflow_lite_wrap_calibration_wrapper")
     29 
     30 
     31 class Calibrator(object):
     32   """Calibrates a floating point model and then quantizes it.
     33 
     34   This is an internal class, not a public interface.
     35   """
     36 
     37   def __init__(self, model_content):
     38     """Constructor.
     39 
     40     Args:
     41       model_content: Content of a TF-Lite Flatbuffer file.
     42 
     43     Raises:
     44       ValueError: If the calibrator was unable to open the model.
     45     """
     46     if not model_content:
     47       raise ValueError("`model_content` must be specified.")
     48     try:
     49       self._calibrator = (_calibration_wrapper.CalibrationWrapper
     50                           .CreateWrapperCPPFromBuffer(model_content))
     51     except Exception as e:
     52       raise ValueError("Failed to parse the model: %s." % e)
     53     if not self._calibrator:
     54       raise ValueError("Failed to parse the model.")
     55 
     56   def calibrate_and_quantize(self, dataset_gen):
     57     """Calibrates the model with specified generator and then quantizes it.
     58 
     59     Returns:
     60       A quantized model.
     61 
     62     Args:
     63       dataset_gen: A generator that generates calibration samples.
     64     """
     65     self._calibrator.Prepare()
     66     for calibration_sample in dataset_gen():
     67       self._calibrator.FeedTensor(calibration_sample)
     68     return self._calibrator.QuantizeModel()
     69