Преобразование модели Jax для TFLite

Обзор

Эта CodeLab демонстрирует, как построить модель для распознавания MNIST с помощью Jax и как преобразовать ее в TensorFlow Lite. Эта лаборатория кода также продемонстрирует, как оптимизировать модель TFLite, преобразованную в Jax, с помощью квантования после обучения.

Посмотреть на TensorFlow.org Запускаем в Google Colab Посмотреть исходный код на GitHub Скачать блокнот

Предпосылки

Рекомендуется попробовать эту функцию с новейшей ночной сборкой TensorFlow.

pip install tf-nightly --upgrade
pip install jax --upgrade
pip install jaxlib --upgrade

Подготовка данных

Загрузите данные MNIST с набором данных Keras и выполните предварительную обработку.

import numpy as np
import tensorflow as tf
import functools

import time
import itertools

import numpy.random as npr

import jax.numpy as jnp
from jax import jit, grad, random
from jax.experimental import optimizers
from jax.experimental import stax
def _one_hot(x, k, dtype=np.float32):
  """Create a one-hot encoding of x of size k."""
  return np.array(x[:, None] == np.arange(k), dtype)

(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()
train_images, test_images = train_images / 255.0, test_images / 255.0
train_images = train_images.astype(np.float32)
test_images = test_images.astype(np.float32)

train_labels = _one_hot(train_labels, 10)
test_labels = _one_hot(test_labels, 10)
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz
11493376/11490434 [==============================] - 0s 0us/step
11501568/11490434 [==============================] - 0s 0us/step

Постройте модель MNIST с помощью Jax

def loss(params, batch):
  inputs, targets = batch
  preds = predict(params, inputs)
  return -jnp.mean(jnp.sum(preds * targets, axis=1))

def accuracy(params, batch):
  inputs, targets = batch
  target_class = jnp.argmax(targets, axis=1)
  predicted_class = jnp.argmax(predict(params, inputs), axis=1)
  return jnp.mean(predicted_class == target_class)

init_random_params, predict = stax.serial(
    stax.Flatten,
    stax.Dense(1024), stax.Relu,
    stax.Dense(1024), stax.Relu,
    stax.Dense(10), stax.LogSoftmax)

rng = random.PRNGKey(0)
WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

Обучите и оцените модель

step_size = 0.001
num_epochs = 10
batch_size = 128
momentum_mass = 0.9


num_train = train_images.shape[0]
num_complete_batches, leftover = divmod(num_train, batch_size)
num_batches = num_complete_batches + bool(leftover)

def data_stream():
  rng = npr.RandomState(0)
  while True:
    perm = rng.permutation(num_train)
    for i in range(num_batches):
      batch_idx = perm[i * batch_size:(i + 1) * batch_size]
      yield train_images[batch_idx], train_labels[batch_idx]
batches = data_stream()

opt_init, opt_update, get_params = optimizers.momentum(step_size, mass=momentum_mass)

@jit
def update(i, opt_state, batch):
  params = get_params(opt_state)
  return opt_update(i, grad(loss)(params, batch), opt_state)

_, init_params = init_random_params(rng, (-1, 28 * 28))
opt_state = opt_init(init_params)
itercount = itertools.count()

print("\nStarting training...")
for epoch in range(num_epochs):
  start_time = time.time()
  for _ in range(num_batches):
    opt_state = update(next(itercount), opt_state, next(batches))
  epoch_time = time.time() - start_time

  params = get_params(opt_state)
  train_acc = accuracy(params, (train_images, train_labels))
  test_acc = accuracy(params, (test_images, test_labels))
  print("Epoch {} in {:0.2f} sec".format(epoch, epoch_time))
  print("Training set accuracy {}".format(train_acc))
  print("Test set accuracy {}".format(test_acc))
Starting training...
Epoch 0 in 4.69 sec
Training set accuracy 0.8729000091552734
Test set accuracy 0.880299985408783
Epoch 1 in 3.83 sec
Training set accuracy 0.8983666896820068
Test set accuracy 0.9047999978065491
Epoch 2 in 3.81 sec
Training set accuracy 0.9102166891098022
Test set accuracy 0.9138000011444092
Epoch 3 in 3.85 sec
Training set accuracy 0.9172500371932983
Test set accuracy 0.9218999743461609
Epoch 4 in 3.79 sec
Training set accuracy 0.9224500060081482
Test set accuracy 0.9253999590873718
Epoch 5 in 3.72 sec
Training set accuracy 0.9272000193595886
Test set accuracy 0.930899977684021
Epoch 6 in 3.77 sec
Training set accuracy 0.9327666759490967
Test set accuracy 0.9334999918937683
Epoch 7 in 3.77 sec
Training set accuracy 0.9360166788101196
Test set accuracy 0.9370999932289124
Epoch 8 in 3.77 sec
Training set accuracy 0.9390000104904175
Test set accuracy 0.939300000667572
Epoch 9 in 3.73 sec
Training set accuracy 0.9425666928291321
Test set accuracy 0.9429999589920044

Преобразование в модель TFLite.

Обратите внимание: мы

  1. Встроенный в PARAMS к Jax predict FUNC с functools.partial .
  2. Построить jnp.zeros , это «заполнитель» Тензор используется для Jax , чтобы проследить модель.
  3. Вызов experimental_from_jax :> * The serving_func заворачивают в списке. > * Входные данные связаны с заданным именем и передаются в виде массива, заключенного в список.
serving_func = functools.partial(predict, params)
x_input = jnp.zeros((1, 28, 28))
converter = tf.lite.TFLiteConverter.experimental_from_jax(
    [serving_func], [[('input1', x_input)]])
tflite_model = converter.convert()
with open('jax_mnist.tflite', 'wb') as f:
  f.write(tflite_model)
2021-10-30 11:51:13.208329: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:363] Ignored output_format.
2021-10-30 11:51:13.208375: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:366] Ignored drop_control_dependency.
2021-10-30 11:51:13.208383: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:372] Ignored change_concat_input_ranges.

Проверьте преобразованную модель TFLite

Сравните результаты преобразованной модели с моделью Jax.

expected = serving_func(train_images[0:1])

# Run the model with TensorFlow Lite
interpreter = tf.lite.Interpreter(model_content=tflite_model)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
interpreter.set_tensor(input_details[0]["index"], train_images[0:1, :, :])
interpreter.invoke()
result = interpreter.get_tensor(output_details[0]["index"])

# Assert if the result of TFLite model is consistent with the JAX model.
np.testing.assert_almost_equal(expected, result, 1e-5)

Оптимизировать модель

Мы обеспечим representative_dataset делать после обучения quantiztion для оптимизации модели.

def representative_dataset():
  for i in range(1000):
    x = train_images[i:i+1]
    yield [x]

converter = tf.lite.TFLiteConverter.experimental_from_jax(
    [serving_func], [[('x', x_input)]])
tflite_model = converter.convert()
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.representative_dataset = representative_dataset
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
tflite_quant_model = converter.convert()
with open('jax_mnist_quant.tflite', 'wb') as f:
  f.write(tflite_quant_model)
2021-10-30 11:51:14.202412: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:363] Ignored output_format.
2021-10-30 11:51:14.202455: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:366] Ignored drop_control_dependency.
2021-10-30 11:51:14.202461: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:372] Ignored change_concat_input_ranges.
2021-10-30 11:51:14.293677: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:363] Ignored output_format.
2021-10-30 11:51:14.293768: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:366] Ignored drop_control_dependency.
2021-10-30 11:51:14.293776: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:372] Ignored change_concat_input_ranges.
fully_quantize: 0, inference_type: 6, input_inference_type: 0, output_inference_type: 0

Оцените оптимизированную модель

expected = serving_func(train_images[0:1])

# Run the model with TensorFlow Lite
interpreter = tf.lite.Interpreter(model_content=tflite_quant_model)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
interpreter.set_tensor(input_details[0]["index"], train_images[0:1, :, :])
interpreter.invoke()
result = interpreter.get_tensor(output_details[0]["index"])

# Assert if the result of TFLite model is consistent with the Jax model.
np.testing.assert_almost_equal(expected, result, 1e-5)

Сравните размер квантованной модели

Мы должны увидеть, что квантованная модель в четыре раза меньше исходной.

du -h jax_mnist.tflite
du -h jax_mnist_quant.tflite
7.2M    jax_mnist.tflite
1.8M    jax_mnist_quant.tflite