Attend the Women in ML Symposium on December 7 Register now

Pruning for on-device inference w/ XNNPACK

Stay organized with collections Save and categorize content based on your preferences.

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

Welcome to the guide on Keras weights pruning for improving latency of on-device inference via XNNPACK.

This guide presents the usage of the newly introduced tfmot.sparsity.keras.PruningPolicy API and demonstrates how it could be used for accelerating mostly convolutional models on modern CPUs using XNNPACK Sparse inference.

The guide covers the following steps of the model creation process:

  • Build and train the dense baseline
  • Fine-tune model with pruning
  • Convert to TFLite
  • On-device benchmark

The guide doesn't cover the best practices for the fine-tuning with pruning. For more detailed information on this topic, please check out our comprehensive guide.

Setup

 pip install -q tensorflow
 pip install -q tensorflow-model-optimization
import tempfile

import tensorflow as tf
import numpy as np

from tensorflow import keras
import tensorflow_datasets as tfds
import tensorflow_model_optimization as tfmot

%load_ext tensorboard
2022-10-20 11:46:57.446616: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory
2022-10-20 11:46:57.446722: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory
2022-10-20 11:46:57.446734: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

Build and train the dense model

We build and train a simple baseline CNN for classification task on CIFAR10 dataset.

# Load CIFAR10 dataset.
(ds_train, ds_val, ds_test), ds_info = tfds.load(
    'cifar10',
    split=['train[:90%]', 'train[90%:]', 'test'],
    as_supervised=True,
    with_info=True,
)

# Normalize the input image so that each pixel value is between 0 and 1.
def normalize_img(image, label):
  """Normalizes images: `uint8` -> `float32`."""
  return tf.image.convert_image_dtype(image, tf.float32), label

# Load the data in batches of 128 images.
batch_size = 128
def prepare_dataset(ds, buffer_size=None):
  ds = ds.map(normalize_img, num_parallel_calls=tf.data.experimental.AUTOTUNE)
  ds = ds.cache()
  if buffer_size:
    ds = ds.shuffle(buffer_size)
  ds = ds.batch(batch_size)
  ds = ds.prefetch(tf.data.experimental.AUTOTUNE)
  return ds

ds_train = prepare_dataset(ds_train,
                           buffer_size=ds_info.splits['train'].num_examples)
ds_val = prepare_dataset(ds_val)
ds_test = prepare_dataset(ds_test)

# Build the dense baseline model.
dense_model = keras.Sequential([
    keras.layers.InputLayer(input_shape=(32, 32, 3)),
    keras.layers.ZeroPadding2D(padding=1),
    keras.layers.Conv2D(
        filters=8,
        kernel_size=(3, 3),
        strides=(2, 2),
        padding='valid'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.DepthwiseConv2D(kernel_size=(3, 3), padding='same'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.Conv2D(filters=16, kernel_size=(1, 1)),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.ZeroPadding2D(padding=1),
    keras.layers.DepthwiseConv2D(
        kernel_size=(3, 3), strides=(2, 2), padding='valid'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.Conv2D(filters=32, kernel_size=(1, 1)),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.GlobalAveragePooling2D(),
    keras.layers.Flatten(),
    keras.layers.Dense(10)
])

# Compile and train the dense model for 10 epochs.
dense_model.compile(
    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    optimizer='adam',
    metrics=['accuracy'])

dense_model.fit(
  ds_train,
  epochs=10,
  validation_data=ds_val)

# Evaluate the dense model.
_, dense_model_accuracy = dense_model.evaluate(ds_test, verbose=0)
2022-10-20 11:46:59.886689: E tensorflow/compiler/xla/stream_executor/cuda/cuda_driver.cc:267] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
Epoch 1/10
352/352 [==============================] - 22s 51ms/step - loss: 2.0083 - accuracy: 0.2677 - val_loss: 2.1805 - val_accuracy: 0.1562
Epoch 2/10
352/352 [==============================] - 6s 16ms/step - loss: 1.7056 - accuracy: 0.3635 - val_loss: 1.6602 - val_accuracy: 0.3696
Epoch 3/10
352/352 [==============================] - 6s 16ms/step - loss: 1.5999 - accuracy: 0.4085 - val_loss: 1.7108 - val_accuracy: 0.3742
Epoch 4/10
352/352 [==============================] - 6s 16ms/step - loss: 1.5433 - accuracy: 0.4349 - val_loss: 1.5007 - val_accuracy: 0.4466
Epoch 5/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4983 - accuracy: 0.4534 - val_loss: 1.5216 - val_accuracy: 0.4372
Epoch 6/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4665 - accuracy: 0.4695 - val_loss: 1.4703 - val_accuracy: 0.4656
Epoch 7/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4442 - accuracy: 0.4767 - val_loss: 1.4679 - val_accuracy: 0.4670
Epoch 8/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4279 - accuracy: 0.4822 - val_loss: 1.6021 - val_accuracy: 0.4210
Epoch 9/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4119 - accuracy: 0.4907 - val_loss: 1.4191 - val_accuracy: 0.4730
Epoch 10/10
352/352 [==============================] - 6s 16ms/step - loss: 1.4013 - accuracy: 0.4945 - val_loss: 1.3932 - val_accuracy: 0.4886

Build the sparse model

Using the instructions from the comprehensive guide, we apply tfmot.sparsity.keras.prune_low_magnitude function with parameters that target on-device acceleration via pruning i.e. tfmot.sparsity.keras.PruneForLatencyOnXNNPack policy.

prune_low_magnitude = tfmot.sparsity.keras.prune_low_magnitude

# Compute end step to finish pruning after after 5 epochs.
end_epoch = 5

num_iterations_per_epoch = len(ds_train)
end_step =  num_iterations_per_epoch * end_epoch

# Define parameters for pruning.
pruning_params = {
      'pruning_schedule': tfmot.sparsity.keras.PolynomialDecay(initial_sparsity=0.25,
                                                               final_sparsity=0.75,
                                                               begin_step=0,
                                                               end_step=end_step),
      'pruning_policy': tfmot.sparsity.keras.PruneForLatencyOnXNNPack()
}

# Try to apply pruning wrapper with pruning policy parameter.
try:
  model_for_pruning = prune_low_magnitude(dense_model, **pruning_params)
except ValueError as e:
  print(e)
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/tensorflow/python/autograph/pyct/static_analysis/liveness.py:83: Analyzer.lamba_check (from tensorflow.python.autograph.pyct.static_analysis.liveness) is deprecated and will be removed after 2023-09-23.
Instructions for updating:
Lambda fuctions will be no more assumed to be used in the statement where they are used, or at least in the same block. https://github.com/tensorflow/tensorflow/issues/56089
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/tensorflow/python/autograph/pyct/static_analysis/liveness.py:83: Analyzer.lamba_check (from tensorflow.python.autograph.pyct.static_analysis.liveness) is deprecated and will be removed after 2023-09-23.
Instructions for updating:
Lambda fuctions will be no more assumed to be used in the statement where they are used, or at least in the same block. https://github.com/tensorflow/tensorflow/issues/56089

The call prune_low_magnitude results in ValueError with the message Could not find a GlobalAveragePooling2D layer with keepdims = True in all output branches. The message indicates that the model isn't supported for pruning with policy tfmot.sparsity.keras.PruneForLatencyOnXNNPack and specifically the layer GlobalAveragePooling2D requires the parameter keepdims = True. Let's fix that and reapply prune_low_magnitude function.

fixed_dense_model = keras.Sequential([
    keras.layers.InputLayer(input_shape=(32, 32, 3)),
    keras.layers.ZeroPadding2D(padding=1),
    keras.layers.Conv2D(
        filters=8,
        kernel_size=(3, 3),
        strides=(2, 2),
        padding='valid'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.DepthwiseConv2D(kernel_size=(3, 3), padding='same'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.Conv2D(filters=16, kernel_size=(1, 1)),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.ZeroPadding2D(padding=1),
    keras.layers.DepthwiseConv2D(
        kernel_size=(3, 3), strides=(2, 2), padding='valid'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.Conv2D(filters=32, kernel_size=(1, 1)),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.GlobalAveragePooling2D(keepdims=True),
    keras.layers.Flatten(),
    keras.layers.Dense(10)
])

# Use the pretrained model for pruning instead of training from scratch.
fixed_dense_model.set_weights(dense_model.get_weights())

# Try to reapply pruning wrapper.
model_for_pruning = prune_low_magnitude(fixed_dense_model, **pruning_params)

Invocation of prune_low_magnitude has finished without any errors meaning that the model is fully supported for the tfmot.sparsity.keras.PruneForLatencyOnXNNPack policy and can be accelerated using XNNPACK Sparse inference.

Fine-tune the sparse model

Following the pruning example, we fine-tune the sparse model using the weights of the dense model. We start fine-tuning of the model with 25% sparsity (25% of the weights are set to zero) and end with 75% sparsity.

logdir = tempfile.mkdtemp()

callbacks = [
  tfmot.sparsity.keras.UpdatePruningStep(),
  tfmot.sparsity.keras.PruningSummaries(log_dir=logdir),
]

model_for_pruning.compile(
    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    optimizer='adam',
    metrics=['accuracy'])

model_for_pruning.fit(
  ds_train,
  epochs=15,
  validation_data=ds_val,
  callbacks=callbacks)

# Evaluate the dense model.
_, pruned_model_accuracy = model_for_pruning.evaluate(ds_test, verbose=0)

print('Dense model test accuracy:', dense_model_accuracy)
print('Pruned model test accuracy:', pruned_model_accuracy)
Epoch 1/15
352/352 [==============================] - 9s 17ms/step - loss: 1.4032 - accuracy: 0.4931 - val_loss: 1.5868 - val_accuracy: 0.4132
Epoch 2/15
352/352 [==============================] - 6s 16ms/step - loss: 1.4221 - accuracy: 0.4855 - val_loss: 1.6649 - val_accuracy: 0.4098
Epoch 3/15
352/352 [==============================] - 6s 16ms/step - loss: 1.4408 - accuracy: 0.4773 - val_loss: 1.5593 - val_accuracy: 0.4354
Epoch 4/15
352/352 [==============================] - 6s 17ms/step - loss: 1.4445 - accuracy: 0.4788 - val_loss: 1.5197 - val_accuracy: 0.4494
Epoch 5/15
352/352 [==============================] - 6s 17ms/step - loss: 1.4277 - accuracy: 0.4839 - val_loss: 1.4436 - val_accuracy: 0.4726
Epoch 6/15
352/352 [==============================] - 6s 16ms/step - loss: 1.4104 - accuracy: 0.4905 - val_loss: 1.4833 - val_accuracy: 0.4582
Epoch 7/15
352/352 [==============================] - 6s 16ms/step - loss: 1.4019 - accuracy: 0.4960 - val_loss: 1.4347 - val_accuracy: 0.4810
Epoch 8/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3927 - accuracy: 0.4993 - val_loss: 1.5085 - val_accuracy: 0.4544
Epoch 9/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3846 - accuracy: 0.5027 - val_loss: 1.4205 - val_accuracy: 0.4802
Epoch 10/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3786 - accuracy: 0.5034 - val_loss: 1.5446 - val_accuracy: 0.4326
Epoch 11/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3750 - accuracy: 0.5056 - val_loss: 1.4204 - val_accuracy: 0.4848
Epoch 12/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3699 - accuracy: 0.5055 - val_loss: 1.4268 - val_accuracy: 0.4638
Epoch 13/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3646 - accuracy: 0.5095 - val_loss: 1.8506 - val_accuracy: 0.3586
Epoch 14/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3612 - accuracy: 0.5104 - val_loss: 1.4045 - val_accuracy: 0.4906
Epoch 15/15
352/352 [==============================] - 6s 16ms/step - loss: 1.3546 - accuracy: 0.5124 - val_loss: 1.5012 - val_accuracy: 0.4434
Dense model test accuracy: 0.49129998683929443
Pruned model test accuracy: 0.4438999891281128

The logs show the progression of sparsity on a per-layer basis.

#docs_infra: no_execute
%tensorboard --logdir={logdir}

After the fine-tuning with pruning, test accuracy demonstrates a modest improvement (43% to 44%) compared to the dense model. Let's compare on-device latency using TFLite benchmark.

Model conversion and benchmarking

To convert the pruned model into TFLite, we need replace the PruneLowMagnitude wrappers with original layers via the strip_pruning function. Also, since the weights of the pruned model (model_for_pruning) are mostly zeros, we may apply an optimization tf.lite.Optimize.EXPERIMENTAL_SPARSITY to efficiently store the resulted TFLite model. This optimization flag is not required for the dense model.

converter = tf.lite.TFLiteConverter.from_keras_model(dense_model)
dense_tflite_model = converter.convert()

_, dense_tflite_file = tempfile.mkstemp('.tflite')
with open(dense_tflite_file, 'wb') as f:
  f.write(dense_tflite_model)

model_for_export = tfmot.sparsity.keras.strip_pruning(model_for_pruning)

converter = tf.lite.TFLiteConverter.from_keras_model(model_for_export)
converter.optimizations = [tf.lite.Optimize.EXPERIMENTAL_SPARSITY]
pruned_tflite_model = converter.convert()

_, pruned_tflite_file = tempfile.mkstemp('.tflite')
with open(pruned_tflite_file, 'wb') as f:
  f.write(pruned_tflite_model)
WARNING:absl:Found untraced functions such as _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op while saving (showing 5 of 6). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: /tmpfs/tmp/tmp1tp5qvfn/assets
INFO:tensorflow:Assets written to: /tmpfs/tmp/tmp1tp5qvfn/assets
2022-10-20 11:49:48.574550: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:362] Ignored output_format.
2022-10-20 11:49:48.574596: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:365] Ignored drop_control_dependency.
WARNING:absl:Found untraced functions such as _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op, _jit_compiled_convolution_op while saving (showing 5 of 5). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: /tmpfs/tmp/tmp_x_w3jbc/assets
INFO:tensorflow:Assets written to: /tmpfs/tmp/tmp_x_w3jbc/assets
2022-10-20 11:49:52.037791: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:362] Ignored output_format.
2022-10-20 11:49:52.037836: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:365] Ignored drop_control_dependency.

Following the instructions of TFLite Model Benchmarking Tool, we build the tool, upload it to the Android device together with dense and pruned TFLite models, and benchmark both models on the device.

! adb shell /data/local/tmp/benchmark_model \
    --graph=/data/local/tmp/dense_model.tflite \
    --use_xnnpack=true \
    --num_runs=100 \
    --num_threads=1
/bin/bash: adb: command not found
! adb shell /data/local/tmp/benchmark_model \
    --graph=/data/local/tmp/pruned_model.tflite \
    --use_xnnpack=true \
    --num_runs=100 \
    --num_threads=1
/bin/bash: adb: command not found

Benchmarks on Pixel 4 resulted in average inference time of 17us for the dense model and 12us for the pruned model. The on-device benchmarks demonstrate a clear 5us or 30% improvements in latency even for such small models. In our experience, larger models based on MobileNetV3 or EfficientNet-lite show similar performance improvements. The speed-up varies based on the relative contribution of 1x1 convolutions to the overall model.

Conclusion

In this tutorial, we show how one may create sparse models for faster on-device performance using the new functionality introduced by the TF MOT API and XNNPack. These sparse models are smaller and faster than their dense counterparts while retaining or even surpassing their quality.

We encourage you to try this new capability which can be particularly important for deploying your models on device.