Sparsity preserving clustering Keras example

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

Overview

This is an end to end example showing the usage of the sparsity preserving clustering API, part of the TensorFlow Model Optimization Toolkit's collaborative optimization pipeline.

Other pages

For an introduction to the pipeline and other available techniques, see the collaborative optimization overview page.

Contents

In the tutorial, you will:

  1. Train a keras model for the MNIST dataset from scratch.
  2. Fine-tune the model with sparsity and see the accuracy and observe that the model was successfully pruned.
  3. Apply weight clustering to the pruned model and observe the loss of sparsity.
  4. Apply sparsity preserving clustering on the pruned model and observe that the sparsity applied earlier has been preserved.
  5. Generate a TFLite model and check that the accuracy has been preserved in the pruned clustered model.
  6. Compare the sizes of the different models to observe the compression benefits of applying sparsity followed by the collaborative optimization technique of sparsity preserving clustering.

Setup

You can run this Jupyter Notebook in your local virtualenv or colab. For details of setting up dependencies, please refer to the installation guide. 

 pip install -q tensorflow-model-optimization
 
import tensorflow as tf
import tf_keras as keras

import numpy as np
import tempfile
import zipfile
import os

Train a keras model for MNIST to be pruned and clustered

# Load MNIST dataset
mnist = keras.datasets.mnist
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

# Normalize the input image so that each pixel value is between 0 to 1.
train_images = train_images / 255.0
test_images  = test_images / 255.0

model = keras.Sequential([
  keras.layers.InputLayer(input_shape=(28, 28)),
  keras.layers.Reshape(target_shape=(28, 28, 1)),
  keras.layers.Conv2D(filters=12, kernel_size=(3, 3),
                         activation=tf.nn.relu),
  keras.layers.MaxPooling2D(pool_size=(2, 2)),
  keras.layers.Flatten(),
  keras.layers.Dense(10)
])

# Train the digit classification model
model.compile(optimizer='adam',
              loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

model.fit(
    train_images,
    train_labels,
    validation_split=0.1,
    epochs=10
)

2024-03-09 12:54:09.347032: E external/local_xla/xla/stream_executor/cuda/cuda_driver.cc:282] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
Epoch 1/10
1688/1688 [==============================] - 21s 4ms/step - loss: 0.3119 - accuracy: 0.9120 - val_loss: 0.1272 - val_accuracy: 0.9640
Epoch 2/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.1224 - accuracy: 0.9655 - val_loss: 0.0870 - val_accuracy: 0.9770
Epoch 3/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0908 - accuracy: 0.9740 - val_loss: 0.0740 - val_accuracy: 0.9800
Epoch 4/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0759 - accuracy: 0.9775 - val_loss: 0.0639 - val_accuracy: 0.9830
Epoch 5/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0659 - accuracy: 0.9810 - val_loss: 0.0653 - val_accuracy: 0.9832
Epoch 6/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0593 - accuracy: 0.9822 - val_loss: 0.0675 - val_accuracy: 0.9805
Epoch 7/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0538 - accuracy: 0.9839 - val_loss: 0.0615 - val_accuracy: 0.9825
Epoch 8/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0489 - accuracy: 0.9848 - val_loss: 0.0619 - val_accuracy: 0.9832
Epoch 9/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0457 - accuracy: 0.9862 - val_loss: 0.0639 - val_accuracy: 0.9838
Epoch 10/10
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0422 - accuracy: 0.9870 - val_loss: 0.0593 - val_accuracy: 0.9835
<tf_keras.src.callbacks.History at 0x7f0f2dac7b80>

Evaluate the baseline model and save it for later usage

_, baseline_model_accuracy = model.evaluate(
    test_images, test_labels, verbose=0)

print('Baseline test accuracy:', baseline_model_accuracy)

_, keras_file = tempfile.mkstemp('.h5')
print('Saving model to: ', keras_file)
keras.models.save_model(model, keras_file, include_optimizer=False)

Baseline test accuracy: 0.98089998960495
Saving model to:  /tmpfs/tmp/tmp98l4xiax.h5
/tmpfs/tmp/ipykernel_44770/3680774635.py:8: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native TF-Keras format, e.g. `model.save('my_model.keras')`.
  keras.models.save_model(model, keras_file, include_optimizer=False)

Prune and fine-tune the model to 50% sparsity

Apply the prune_low_magnitude() API to prune the whole pre-trained model to achieve the model that is to be clustered in the next step. For how best to use the API to achieve the best compression rate while maintaining your target accuracy, refer to the pruning comprehensive guide.

Define the model and apply the sparsity API

Note that the pre-trained model is used.

import tensorflow_model_optimization as tfmot

prune_low_magnitude = tfmot.sparsity.keras.prune_low_magnitude

pruning_params = {
      'pruning_schedule': tfmot.sparsity.keras.ConstantSparsity(0.5, begin_step=0, frequency=100)
  }

callbacks = [
  tfmot.sparsity.keras.UpdatePruningStep()
]

pruned_model = prune_low_magnitude(model, **pruning_params)

# Use smaller learning rate for fine-tuning
opt = keras.optimizers.Adam(learning_rate=1e-5)

pruned_model.compile(
  loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
  optimizer=opt,
  metrics=['accuracy'])

pruned_model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 prune_low_magnitude_reshap  (None, 28, 28, 1)         1         
 e (PruneLowMagnitude)                                           
                                                                 
 prune_low_magnitude_conv2d  (None, 26, 26, 12)        230       
  (PruneLowMagnitude)                                            
                                                                 
 prune_low_magnitude_max_po  (None, 13, 13, 12)        1         
 oling2d (PruneLowMagnitude                                      
 )                                                               
                                                                 
 prune_low_magnitude_flatte  (None, 2028)              1         
 n (PruneLowMagnitude)                                           
                                                                 
 prune_low_magnitude_dense   (None, 10)                40572     
 (PruneLowMagnitude)                                             
                                                                 
=================================================================
Total params: 40805 (159.41 KB)
Trainable params: 20410 (79.73 KB)
Non-trainable params: 20395 (79.69 KB)
_________________________________________________________________

Fine-tune the model, check sparsity, and evaluate the accuracy against baseline

Fine-tune the model with pruning for 3 epochs.

# Fine-tune model
pruned_model.fit(
  train_images,
  train_labels,
  epochs=3,
  validation_split=0.1,
  callbacks=callbacks)

Epoch 1/3
1688/1688 [==============================] - 10s 4ms/step - loss: 0.0805 - accuracy: 0.9729 - val_loss: 0.0834 - val_accuracy: 0.9753
Epoch 2/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0636 - accuracy: 0.9791 - val_loss: 0.0735 - val_accuracy: 0.9798
Epoch 3/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0557 - accuracy: 0.9825 - val_loss: 0.0688 - val_accuracy: 0.9813
<tf_keras.src.callbacks.History at 0x7f0f2d9bc220>

Define helper functions to calculate and print the sparsity of the model.

def print_model_weights_sparsity(model):

    for layer in model.layers:
        if isinstance(layer, keras.layers.Wrapper):
            weights = layer.trainable_weights
        else:
            weights = layer.weights
        for weight in weights:
            if "kernel" not in weight.name or "centroid" in weight.name:
                continue
            weight_size = weight.numpy().size
            zero_num = np.count_nonzero(weight == 0)
            print(
                f"{weight.name}: {zero_num/weight_size:.2%} sparsity ",
                f"({zero_num}/{weight_size})",
            )

Check that the model kernels was correctly pruned. We need to strip the pruning wrapper first. We also create a deep copy of the model to be used in the next step.

stripped_pruned_model = tfmot.sparsity.keras.strip_pruning(pruned_model)

print_model_weights_sparsity(stripped_pruned_model)

stripped_pruned_model_copy = keras.models.clone_model(stripped_pruned_model)
stripped_pruned_model_copy.set_weights(stripped_pruned_model.get_weights())

conv2d/kernel:0: 50.00% sparsity  (54/108)
dense/kernel:0: 50.00% sparsity  (10140/20280)

Apply clustering and sparsity preserving clustering and check its effect on model sparsity in both cases

Next, we apply both clustering and sparsity preserving clustering on the pruned model and observe that the latter preserves sparsity on your pruned model. Note that we stripped pruning wrappers from the pruned model with tfmot.sparsity.keras.strip_pruning before applying the clustering API.

# Clustering
cluster_weights = tfmot.clustering.keras.cluster_weights
CentroidInitialization = tfmot.clustering.keras.CentroidInitialization

clustering_params = {
  'number_of_clusters': 8,
  'cluster_centroids_init': CentroidInitialization.KMEANS_PLUS_PLUS
}

clustered_model = cluster_weights(stripped_pruned_model, **clustering_params)

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

print('Train clustering model:')
clustered_model.fit(train_images, train_labels,epochs=3, validation_split=0.1)


stripped_pruned_model.save("stripped_pruned_model_clustered.h5")

# Sparsity preserving clustering
from tensorflow_model_optimization.python.core.clustering.keras.experimental import (
    cluster,
)

cluster_weights = cluster.cluster_weights

clustering_params = {
  'number_of_clusters': 8,
  'cluster_centroids_init': CentroidInitialization.KMEANS_PLUS_PLUS,
  'preserve_sparsity': True
}

sparsity_clustered_model = cluster_weights(stripped_pruned_model_copy, **clustering_params)

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

print('Train sparsity preserving clustering model:')
sparsity_clustered_model.fit(train_images, train_labels,epochs=3, validation_split=0.1)

Train clustering model:
Epoch 1/3
1688/1688 [==============================] - 9s 5ms/step - loss: 0.0477 - accuracy: 0.9846 - val_loss: 0.0659 - val_accuracy: 0.9813
Epoch 2/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0464 - accuracy: 0.9851 - val_loss: 0.0611 - val_accuracy: 0.9825
Epoch 3/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0452 - accuracy: 0.9855 - val_loss: 0.0728 - val_accuracy: 0.9797
WARNING:tensorflow:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
/tmpfs/src/tf_docs_env/lib/python3.9/site-packages/tf_keras/src/engine/training.py:3098: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native TF-Keras format, e.g. `model.save('my_model.keras')`.
  saving_api.save_model(
Train sparsity preserving clustering model:
Epoch 1/3
1688/1688 [==============================] - 9s 5ms/step - loss: 0.0471 - accuracy: 0.9853 - val_loss: 0.0669 - val_accuracy: 0.9823
Epoch 2/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0446 - accuracy: 0.9863 - val_loss: 0.0661 - val_accuracy: 0.9817
Epoch 3/3
1688/1688 [==============================] - 7s 4ms/step - loss: 0.0435 - accuracy: 0.9857 - val_loss: 0.0714 - val_accuracy: 0.9813
<tf_keras.src.callbacks.History at 0x7f0e4415cc40>

Check sparsity for both models.

print("Clustered Model sparsity:\n")
print_model_weights_sparsity(clustered_model)
print("\nSparsity preserved clustered Model sparsity:\n")
print_model_weights_sparsity(sparsity_clustered_model)

Clustered Model sparsity:

conv2d/kernel:0: 0.00% sparsity  (0/108)
dense/kernel:0: 0.34% sparsity  (69/20280)

Sparsity preserved clustered Model sparsity:

conv2d/kernel:0: 50.00% sparsity  (54/108)
dense/kernel:0: 50.00% sparsity  (10140/20280)

Create 1.6x smaller models from clustering

Define helper function to get zipped model file.

def get_gzipped_model_size(file):
  # It returns the size of the gzipped model in kilobytes.

  _, zipped_file = tempfile.mkstemp('.zip')
  with zipfile.ZipFile(zipped_file, 'w', compression=zipfile.ZIP_DEFLATED) as f:
    f.write(file)

  return os.path.getsize(zipped_file)/1000

# Clustered model
clustered_model_file = 'clustered_model.h5'

# Save the model.
clustered_model.save(clustered_model_file)

#Sparsity Preserve Clustered model
sparsity_clustered_model_file = 'sparsity_clustered_model.h5'

# Save the model.
sparsity_clustered_model.save(sparsity_clustered_model_file)

print("Clustered Model size: ", get_gzipped_model_size(clustered_model_file), ' KB')
print("Sparsity preserved clustered Model size: ", get_gzipped_model_size(sparsity_clustered_model_file), ' KB')

Clustered Model size:  247.191  KB
Sparsity preserved clustered Model size:  155.272  KB

Create a TFLite model from combining sparsity preserving weight clustering and post-training quantization

Strip clustering wrappers and convert to TFLite.

stripped_sparsity_clustered_model = tfmot.clustering.keras.strip_clustering(sparsity_clustered_model)

converter = tf.lite.TFLiteConverter.from_keras_model(stripped_sparsity_clustered_model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
sparsity_clustered_quant_model = converter.convert()

_, pruned_and_clustered_tflite_file = tempfile.mkstemp('.tflite')

with open(pruned_and_clustered_tflite_file, 'wb') as f:
  f.write(sparsity_clustered_quant_model)

print("Sparsity preserved clustered Model size: ", get_gzipped_model_size(sparsity_clustered_model_file), ' KB')
print("Sparsity preserved clustered and quantized TFLite model size:",
       get_gzipped_model_size(pruned_and_clustered_tflite_file), ' KB')

INFO:tensorflow:Assets written to: /tmpfs/tmp/tmpzf7jus7v/assets
INFO:tensorflow:Assets written to: /tmpfs/tmp/tmpzf7jus7v/assets
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
W0000 00:00:1709989008.133294   44770 tf_tfl_flatbuffer_helpers.cc:390] Ignored output_format.
W0000 00:00:1709989008.133348   44770 tf_tfl_flatbuffer_helpers.cc:393] Ignored drop_control_dependency.
Sparsity preserved clustered Model size:  155.272  KB
Sparsity preserved clustered and quantized TFLite model size: 8.183  KB

See the persistence of accuracy from TF to TFLite

def eval_model(interpreter):
  input_index = interpreter.get_input_details()[0]["index"]
  output_index = interpreter.get_output_details()[0]["index"]

  # Run predictions on every image in the "test" dataset.
  prediction_digits = []
  for i, test_image in enumerate(test_images):
    if i % 1000 == 0:
      print(f"Evaluated on {i} results so far.")
    # Pre-processing: add batch dimension and convert to float32 to match with
    # the model's input data format.
    test_image = np.expand_dims(test_image, axis=0).astype(np.float32)
    interpreter.set_tensor(input_index, test_image)

    # Run inference.
    interpreter.invoke()

    # Post-processing: remove batch dimension and find the digit with highest
    # probability.
    output = interpreter.tensor(output_index)
    digit = np.argmax(output()[0])
    prediction_digits.append(digit)

  print('\n')
  # Compare prediction results with ground truth labels to calculate accuracy.
  prediction_digits = np.array(prediction_digits)
  accuracy = (prediction_digits == test_labels).mean()
  return accuracy

You evaluate the model, which has been pruned, clustered and quantized, and then see that the accuracy from TensorFlow persists in the TFLite backend.

# Keras model evaluation
stripped_sparsity_clustered_model.compile(optimizer='adam',
              loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
_, sparsity_clustered_keras_accuracy = stripped_sparsity_clustered_model.evaluate(
    test_images, test_labels, verbose=0)

# TFLite model evaluation
interpreter = tf.lite.Interpreter(pruned_and_clustered_tflite_file)
interpreter.allocate_tensors()

sparsity_clustered_tflite_accuracy = eval_model(interpreter)

print('Pruned, clustered and quantized Keras model accuracy:', sparsity_clustered_keras_accuracy)
print('Pruned, clustered and quantized TFLite model accuracy:', sparsity_clustered_tflite_accuracy)

Evaluated on 0 results so far.
Evaluated on 1000 results so far.
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
WARNING: Attempting to use a delegate that only supports static-sized tensors with a graph that has dynamic-sized tensors (tensor#13 is a dynamic-sized tensor).
Evaluated on 2000 results so far.
Evaluated on 3000 results so far.
Evaluated on 4000 results so far.
Evaluated on 5000 results so far.
Evaluated on 6000 results so far.
Evaluated on 7000 results so far.
Evaluated on 8000 results so far.
Evaluated on 9000 results so far.


Pruned, clustered and quantized Keras model accuracy: 0.9782999753952026
Pruned, clustered and quantized TFLite model accuracy: 0.9784

Conclusion

In this tutorial, you learned how to create a model, prune it using the prune_low_magnitude() API, and apply sparsity preserving clustering to preserve sparsity while clustering the weights. The sparsity preserving clustered model was compared to a clustered one to show that sparsity is preserved in the former and lost in the latter. Next, the pruned clustered model was converted to TFLite to show the compression benefits of chaining the pruning and sparsity preserving clustering model optimization techniques and, finally, the TFLite model was evaluated to ensure that the accuracy persists in the TFLite backend.