Copyright 2019 The TensorFlow Authors.



In [0]:

    
#@title Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

Post-training float16 quantization

View on TensorFlow.org

View source on GitHub

Overview

TensorFlow Lite now supports converting weights to 16-bit floating point values during model conversion from TensorFlow to TensorFlow Lite's flat buffer format. This results in a 2x reduction in model size. Some harware, like GPUs, can compute natively in this reduced precision arithmetic, realizing a speedup over traditional floating point execution. The Tensorflow Lite GPU delegate can be configured to run in this way. However, a model converted to float16 weights can still run on the CPU without additional modification: the float16 weights are upsampled to float32 prior to the first inference. This permits a significant reduction in model size in exchange for a minimal impacts to latency and accuracy.

In this tutorial, you train an MNIST model from scratch, check its accuracy in TensorFlow, and then convert the saved model into a Tensorflow Lite flatbuffer with float16 quantization. Finally, check the accuracy of the converted model and compare it to the original saved model. The training script, mnist.py, is available from the TensorFlow official MNIST tutorial.

Build an MNIST model

Setup



In [0]:

    
! pip uninstall -y tensorflow
! pip install -U tf-nightly



In [0]:

    
import tensorflow as tf
tf.enable_eager_execution()

import numpy as np

tf.logging.set_verbosity(tf.logging.DEBUG)



In [0]:

    
! git clone --depth 1 https://github.com/tensorflow/models



In [0]:

    
tf.lite.constants.FLOAT16



In [0]:

    
import sys
import os

if sys.version_info.major >= 3:
    import pathlib
else:
    import pathlib2 as pathlib

# Add `models` to the python path.
models_path = os.path.join(os.getcwd(), "models")
sys.path.append(models_path)

Train and export the model



In [0]:

    
saved_models_root = "/tmp/mnist_saved_model"



In [0]:

    
# The above path addition is not visible to subprocesses, add the path for the subprocess as well.
!PYTHONPATH={models_path} python models/official/mnist/mnist.py --train_epochs=1 --export_dir {saved_models_root} --data_format=channels_last

For the example, you trained the model for just a single epoch, so it only trains to ~96% accuracy.

Convert to a TensorFlow Lite model

The savedmodel directory is named with a timestamp. Select the most recent one:



In [0]:

    
saved_model_dir = str(sorted(pathlib.Path(saved_models_root).glob("*"))[-1])
saved_model_dir

Using the Python TFLiteConverter, the saved model can be converted into a TensorFlow Lite model.

First load the model using the TFLiteConverter:



In [0]:

    
converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)
tflite_model = converter.convert()

Write it out to a .tflite file:



In [0]:

    
tflite_models_dir = pathlib.Path("/tmp/mnist_tflite_models/")
tflite_models_dir.mkdir(exist_ok=True, parents=True)



In [0]:

    
tflite_model_file = tflite_models_dir/"mnist_model.tflite"
tflite_model_file.write_bytes(tflite_model)

To instead quantize the model to float16 on export, first set the optimizations flag to use default optimizations. Then specify that float16 is the supported type on the target platform:



In [0]:

    
tf.logging.set_verbosity(tf.logging.INFO)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.target_spec.supported_types = [tf.lite.constants.FLOAT16]

Finally, convert the model like usual. Note, by default the converted model will still use float input and outputs for invocation convenience.



In [0]:

    
tflite_fp16_model = converter.convert()
tflite_model_fp16_file = tflite_models_dir/"mnist_model_quant_f16.tflite"
tflite_model_fp16_file.write_bytes(tflite_fp16_model)

Note how the resulting file is approximately 1/2 the size.



In [0]:

    
!ls -lh {tflite_models_dir}

Run the TensorFlow Lite models

Run the TensorFlow Lite model using the Python TensorFlow Lite Interpreter.

Load the test data

First, let's load the MNIST test data to feed to the model:



In [0]:

    
_, mnist_test = tf.keras.datasets.mnist.load_data()
images, labels = tf.cast(mnist_test[0], tf.float32)/255.0, mnist_test[1]

mnist_ds = tf.data.Dataset.from_tensor_slices((images, labels)).batch(1)

Load the model into the interpreters



In [0]:

    
interpreter = tf.lite.Interpreter(model_path=str(tflite_model_file))
interpreter.allocate_tensors()



In [0]:

    
interpreter_fp16 = tf.lite.Interpreter(model_path=str(tflite_model_fp16_file))
interpreter_fp16.allocate_tensors()

Test the models on one image



In [0]:

    
for img, label in mnist_ds:
  break

interpreter.set_tensor(interpreter.get_input_details()[0]["index"], img)
interpreter.invoke()
predictions = interpreter.get_tensor(
    interpreter.get_output_details()[0]["index"])



In [0]:

    
import matplotlib.pylab as plt

plt.imshow(img[0])
template = "True:{true}, predicted:{predict}"
_ = plt.title(template.format(true= str(label[0].numpy()),
                              predict=str(predictions[0])))
plt.grid(False)



In [0]:

    
interpreter_fp16.set_tensor(
    interpreter_fp16.get_input_details()[0]["index"], img)
interpreter_fp16.invoke()
predictions = interpreter_fp16.get_tensor(
    interpreter_fp16.get_output_details()[0]["index"])



In [0]:

    
plt.imshow(img[0])
template = "True:{true}, predicted:{predict}"
_ = plt.title(template.format(true= str(label[0].numpy()),
                              predict=str(predictions[0])))
plt.grid(False)

Evaluate the models



In [0]:

    
def eval_model(interpreter, mnist_ds):
  total_seen = 0
  num_correct = 0

  input_index = interpreter.get_input_details()[0]["index"]
  output_index = interpreter.get_output_details()[0]["index"]
  for img, label in mnist_ds:
    total_seen += 1
    interpreter.set_tensor(input_index, img)
    interpreter.invoke()
    predictions = interpreter.get_tensor(output_index)
    if predictions == label.numpy():
      num_correct += 1

    if total_seen % 500 == 0:
      print("Accuracy after %i images: %f" %
            (total_seen, float(num_correct) / float(total_seen)))

  return float(num_correct) / float(total_seen)



In [0]:

    
# Create smaller dataset for demonstration purposes
mnist_ds_demo = mnist_ds.take(2000)

print(eval_model(interpreter, mnist_ds_demo))

Repeat the evaluation on the float16 quantized model to obtain:



In [0]:

    
# NOTE: Colab runs on server CPUs. At the time of writing this, TensorFlow Lite
# doesn't have super optimized server CPU kernels. For this reason this may be
# slower than the above float interpreter. But for mobile CPUs, considerable
# speedup can be observed.
print(eval_model(interpreter_fp16, mnist_ds_demo))

In this example, you have quantized a model to float16 with no difference in the accuracy.

It's also possible to evaluate the fp16 quantized model on the GPU. To perform all arithmetic with the reduced precision values, be sure to create the TfLiteGPUDelegateOptions struct in your app and set precision_loss_allowed to 1, like this:

//Prepare GPU delegate.
const TfLiteGpuDelegateOptions options = {
  .metadata = NULL,
  .compile_options = {
    .precision_loss_allowed = 1,  // FP16
    .preferred_gl_object_type = TFLITE_GL_OBJECT_TYPE_FASTEST,
    .dynamic_batch_enabled = 0,   // Not fully functional yet
  },
};

Detailed documentation on the TFLite GPU delegate and how to use it in your application can be found here