Questa pagina è stata tradotta dall'API Cloud Translation.
Switch to English

Potatura guida completa

Visualizza su TensorFlow.org Esegui in Google Colab Visualizza sorgente su GitHub Scarica notebook

Benvenuti nella guida completa per la potatura a peso Keras.

Questa pagina documenta vari casi d'uso e mostra come utilizzare l'API per ciascuno di essi. Una volta che sai di quali API hai bisogno, trova i parametri e i dettagli di basso livello nei documenti dell'API .

  • Se vuoi vedere i vantaggi della potatura e cosa è supportato, guarda la panoramica .
  • Per un singolo esempio end-to-end, vedere l' esempio di potatura .

Vengono trattati i seguenti casi d'uso:

  • Definisci e addestra un modello sfoltito.
    • Sequenziale e Funzionale.
    • Modello Keras. Fit e loop di allenamento personalizzati
  • Checkpoint e deserializzazione di un modello sfoltito.
  • Distribuisci un modello ridotto e scopri i vantaggi della compressione.

Per la configurazione dell'algoritmo di tfmot.sparsity.keras.prune_low_magnitude , fare riferimento alla documentazione dell'API tfmot.sparsity.keras.prune_low_magnitude .

Impostare

Per trovare le API di cui hai bisogno e capire gli scopi, puoi eseguire ma saltare la lettura di questa sezione.

! pip install -q tensorflow-model-optimization

import tensorflow as tf
import numpy as np
import tensorflow_model_optimization as tfmot

%load_ext tensorboard

import tempfile

input_shape = [20]
x_train = np.random.randn(1, 20).astype(np.float32)
y_train = tf.keras.utils.to_categorical(np.random.randn(1), num_classes=20)

def setup_model():
  model = tf.keras.Sequential([
      tf.keras.layers.Dense(20, input_shape=input_shape),
      tf.keras.layers.Flatten()
  ])
  return model

def setup_pretrained_weights():
  model = setup_model()

  model.compile(
      loss=tf.keras.losses.categorical_crossentropy,
      optimizer='adam',
      metrics=['accuracy']
  )

  model.fit(x_train, y_train)

  _, pretrained_weights = tempfile.mkstemp('.tf')

  model.save_weights(pretrained_weights)

  return pretrained_weights

def get_gzipped_model_size(model):
  # Returns size of gzipped model, in bytes.
  import os
  import zipfile

  _, keras_file = tempfile.mkstemp('.h5')
  model.save(keras_file, include_optimizer=False)

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

  return os.path.getsize(zipped_file)

setup_model()
pretrained_weights = setup_pretrained_weights()
WARNING: You are using pip version 20.2.2; however, version 20.2.3 is available.
You should consider upgrading via the '/tmpfs/src/tf_docs_env/bin/python -m pip install --upgrade pip' command.
1/1 [==============================] - 0s 2ms/step - loss: 1.1999 - accuracy: 0.0000e+00

Definisci il modello

Potare l'intero modello (sequenziale e funzionale)

Suggerimenti per una migliore precisione del modello:

  • Prova "Elimina alcuni livelli" per saltare l'eliminazione dei livelli che riducono maggiormente la precisione.
  • In genere è meglio perfezionare la potatura piuttosto che allenarsi da zero.

Per formare l'intero modello con la potatura, applica tfmot.sparsity.keras.prune_low_magnitude al modello.

base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended.

model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model)

model_for_pruning.summary()
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow_model_optimization/python/core/sparsity/keras/pruning_wrapper.py:200: Layer.add_variable (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.
Instructions for updating:
Please use `layer.add_weight` method instead.
Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_dense_2  (None, 20)                822       
_________________________________________________________________
prune_low_magnitude_flatten_ (None, 20)                1         
=================================================================
Total params: 823
Trainable params: 420
Non-trainable params: 403
_________________________________________________________________

Potare alcuni strati (sequenziali e funzionali)

La potatura di un modello può avere un effetto negativo sulla precisione. Puoi eliminare selettivamente i livelli di un modello per esplorare il compromesso tra precisione, velocità e dimensioni del modello.

Suggerimenti per una migliore precisione del modello:

  • In genere è meglio perfezionare la potatura piuttosto che allenarsi da zero.
  • Prova a potare gli strati successivi invece dei primi.
  • Evitare di potare gli strati critici (ad es. Meccanismo di attenzione).

Altro :

Nell'esempio seguente, elimina solo gli strati Dense .

# Create a base model
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy

# Helper function uses `prune_low_magnitude` to make only the 
# Dense layers train with pruning.
def apply_pruning_to_dense(layer):
  if isinstance(layer, tf.keras.layers.Dense):
    return tfmot.sparsity.keras.prune_low_magnitude(layer)
  return layer

# Use `tf.keras.models.clone_model` to apply `apply_pruning_to_dense` 
# to the layers of the model.
model_for_pruning = tf.keras.models.clone_model(
    base_model,
    clone_function=apply_pruning_to_dense,
)

model_for_pruning.summary()
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_dense_3  (None, 20)                822       
_________________________________________________________________
flatten_3 (Flatten)          (None, 20)                0         
=================================================================
Total params: 822
Trainable params: 420
Non-trainable params: 402
_________________________________________________________________

Sebbene questo esempio abbia utilizzato il tipo di livello per decidere cosa eliminare, il modo più semplice per eliminare un determinato livello è impostare la sua proprietà name e cercare quel nome nella funzione clone_function .

print(base_model.layers[0].name)
dense_3

Precisione del modello più leggibile ma potenzialmente inferiore

Ciò non è compatibile con la messa a punto con la potatura, motivo per cui potrebbe essere meno accurata rispetto agli esempi precedenti che supportano la regolazione fine.

Sebbene prune_low_magnitude possa essere applicato durante la definizione del modello iniziale, il caricamento dei pesi dopo non funziona negli esempi seguenti.

Esempio funzionale

# Use `prune_low_magnitude` to make the `Dense` layer train with pruning.
i = tf.keras.Input(shape=(20,))
x = tfmot.sparsity.keras.prune_low_magnitude(tf.keras.layers.Dense(10))(i)
o = tf.keras.layers.Flatten()(x)
model_for_pruning = tf.keras.Model(inputs=i, outputs=o)

model_for_pruning.summary()
Model: "functional_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 20)]              0         
_________________________________________________________________
prune_low_magnitude_dense_4  (None, 10)                412       
_________________________________________________________________
flatten_4 (Flatten)          (None, 10)                0         
=================================================================
Total params: 412
Trainable params: 210
Non-trainable params: 202
_________________________________________________________________

Esempio sequenziale

# Use `prune_low_magnitude` to make the `Dense` layer train with pruning.
model_for_pruning = tf.keras.Sequential([
  tfmot.sparsity.keras.prune_low_magnitude(tf.keras.layers.Dense(20, input_shape=input_shape)),
  tf.keras.layers.Flatten()
])

model_for_pruning.summary()
Model: "sequential_4"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_dense_5  (None, 20)                822       
_________________________________________________________________
flatten_5 (Flatten)          (None, 20)                0         
=================================================================
Total params: 822
Trainable params: 420
Non-trainable params: 402
_________________________________________________________________

Potare il livello Keras personalizzato o modificare parti del livello da sfoltire

Errore comune: la potatura del bias di solito danneggia troppo la precisione del modello.

tfmot.sparsity.keras.PrunableLayer serve due casi d'uso:

  1. Potare uno strato Keras personalizzato
  2. Modifica parti di un livello Keras integrato per potare.

Ad esempio, per impostazione predefinita l'API elimina solo il kernel del livello Dense . L'esempio seguente elimina anche il bias.

class MyDenseLayer(tf.keras.layers.Dense, tfmot.sparsity.keras.PrunableLayer):

  def get_prunable_weights(self):
    # Prune bias also, though that usually harms model accuracy too much.
    return [self.kernel, self.bias]

# Use `prune_low_magnitude` to make the `MyDenseLayer` layer train with pruning.
model_for_pruning = tf.keras.Sequential([
  tfmot.sparsity.keras.prune_low_magnitude(MyDenseLayer(20, input_shape=input_shape)),
  tf.keras.layers.Flatten()
])

model_for_pruning.summary()
Model: "sequential_5"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_my_dense (None, 20)                843       
_________________________________________________________________
flatten_6 (Flatten)          (None, 20)                0         
=================================================================
Total params: 843
Trainable params: 420
Non-trainable params: 423
_________________________________________________________________

Modello di treno

Model.fit

Chiama il callback tfmot.sparsity.keras.UpdatePruningStep durante l'addestramento.

Per facilitare l'addestramento al debug, utilizzare il callback tfmot.sparsity.keras.PruningSummaries .

# Define the model.
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy
model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model)

log_dir = tempfile.mkdtemp()
callbacks = [
    tfmot.sparsity.keras.UpdatePruningStep(),
    # Log sparsity and other metrics in Tensorboard.
    tfmot.sparsity.keras.PruningSummaries(log_dir=log_dir)
]

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

model_for_pruning.fit(
    x_train,
    y_train,
    callbacks=callbacks,
    epochs=2,
)

%tensorboard --logdir={log_dir}
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
Epoch 1/2
1/1 [==============================] - 0s 3ms/step - loss: 1.2485 - accuracy: 0.0000e+00
Epoch 2/2
WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow/python/ops/summary_ops_v2.py:1277: stop (from tensorflow.python.eager.profiler) is deprecated and will be removed after 2020-07-01.
Instructions for updating:
use `tf.profiler.experimental.stop` instead.
1/1 [==============================] - 0s 2ms/step - loss: 1.1999 - accuracy: 0.0000e+00

Per gli utenti non Colab, è possibile visualizzarei risultati di una precedente esecuzione di questo blocco di codice su TensorBoard.dev .

Ciclo di allenamento personalizzato

Chiama il callback tfmot.sparsity.keras.UpdatePruningStep durante l'addestramento.

Per facilitare l'addestramento al debug, utilizzare il callback tfmot.sparsity.keras.PruningSummaries .

# Define the model.
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy
model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model)

# Boilerplate
loss = tf.keras.losses.categorical_crossentropy
optimizer = tf.keras.optimizers.Adam()
log_dir = tempfile.mkdtemp()
unused_arg = -1
epochs = 2
batches = 1 # example is hardcoded so that the number of batches cannot change.

# Non-boilerplate.
model_for_pruning.optimizer = optimizer
step_callback = tfmot.sparsity.keras.UpdatePruningStep()
step_callback.set_model(model_for_pruning)
log_callback = tfmot.sparsity.keras.PruningSummaries(log_dir=log_dir) # Log sparsity and other metrics in Tensorboard.
log_callback.set_model(model_for_pruning)

step_callback.on_train_begin() # run pruning callback
for _ in range(epochs):
  log_callback.on_epoch_begin(epoch=unused_arg) # run pruning callback
  for _ in range(batches):
    step_callback.on_train_batch_begin(batch=unused_arg) # run pruning callback

    with tf.GradientTape() as tape:
      logits = model_for_pruning(x_train, training=True)
      loss_value = loss(y_train, logits)
      grads = tape.gradient(loss_value, model_for_pruning.trainable_variables)
      optimizer.apply_gradients(zip(grads, model_for_pruning.trainable_variables))

  step_callback.on_epoch_end(batch=unused_arg) # run pruning callback

%tensorboard --logdir={log_dir}
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.

Per gli utenti non Colab, è possibile visualizzarei risultati di una precedente esecuzione di questo blocco di codice su TensorBoard.dev .

Migliora la precisione del modello sfoltito

Per prima cosa, guarda la documentazione dell'API tfmot.sparsity.keras.prune_low_magnitude per capire cos'è una pianificazione di tfmot.sparsity.keras.prune_low_magnitude e la matematica di ciascun tipo di pianificazione di eliminazione.

Suggerimenti :

  • Avere un tasso di apprendimento che non sia né troppo alto né troppo basso durante la potatura del modello. Considera il programma di potatura un iperparametro.

  • Come test rapido, prova a sperimentare l'eliminazione di un modello alla scarsità finale all'inizio dell'addestramento impostando begin_step su 0 con un programma tfmot.sparsity.keras.ConstantSparsity . Potresti essere fortunato con buoni risultati.

  • Non potare molto frequentemente per dare al modello il tempo di riprendersi. Il programma di potatura fornisce una frequenza predefinita decente.

  • Per idee generali per migliorare l'accuratezza del modello, cerca i suggerimenti per i tuoi casi d'uso in "Definisci modello".

Checkpoint e deserializzazione

È necessario mantenere il passaggio dell'ottimizzatore durante il checkpoint. Ciò significa che mentre puoi utilizzare i modelli Keras HDF5 per il checkpoint, non puoi utilizzare i pesi Keras HDF5.

# Define the model.
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy
model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model)

_, keras_model_file = tempfile.mkstemp('.h5')

# Checkpoint: saving the optimizer is necessary (include_optimizer=True is the default).
model_for_pruning.save(keras_model_file, include_optimizer=True)
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.

Quanto sopra si applica in generale. Il codice seguente è necessario solo per il formato del modello HDF5 (non per i pesi HDF5 e altri formati).

# Deserialize model.
with tfmot.sparsity.keras.prune_scope():
  loaded_model = tf.keras.models.load_model(keras_model_file)

loaded_model.summary()
WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.
Model: "sequential_8"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_dense_8  (None, 20)                822       
_________________________________________________________________
prune_low_magnitude_flatten_ (None, 20)                1         
=================================================================
Total params: 823
Trainable params: 420
Non-trainable params: 403
_________________________________________________________________

Distribuisci il modello sfoltito

Esporta modello con compressione delle dimensioni

Errore comune : sia strip_pruning che l'applicazione di un algoritmo di compressione standard (ad esempio tramite gzip) sono necessari per vedere i vantaggi di compressione della potatura.

# Define the model.
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy
model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model)

# Typically you train the model here.

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

print("final model")
model_for_export.summary()

print("\n")
print("Size of gzipped pruned model without stripping: %.2f bytes" % (get_gzipped_model_size(model_for_pruning)))
print("Size of gzipped pruned model with stripping: %.2f bytes" % (get_gzipped_model_size(model_for_export)))
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
final model
Model: "sequential_9"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_9 (Dense)              (None, 20)                420       
_________________________________________________________________
flatten_10 (Flatten)         (None, 20)                0         
=================================================================
Total params: 420
Trainable params: 420
Non-trainable params: 0
_________________________________________________________________


Size of gzipped pruned model without stripping: 3299.00 bytes
Size of gzipped pruned model with stripping: 2876.00 bytes

Ottimizzazioni specifiche dell'hardware

Una volta che diversi backend abilitano l'eliminazione per migliorare la latenza , l'utilizzo della scarsità dei blocchi può migliorare la latenza per determinati hardware.

L'aumento della dimensione del blocco ridurrà la scarsità di picco ottenibile per la precisione di un modello di destinazione. Nonostante ciò, la latenza può ancora migliorare.

Per i dettagli su cosa è supportato per la scarsità dei blocchi, consulta la documentazione dell'API tfmot.sparsity.keras.prune_low_magnitude .

base_model = setup_model()

# For using intrinsics on a CPU with 128-bit registers, together with 8-bit
# quantized weights, a 1x16 block size is nice because the block perfectly
# fits into the register.
pruning_params = {'block_size': [1, 16]}
model_for_pruning = tfmot.sparsity.keras.prune_low_magnitude(base_model, **pruning_params)

model_for_pruning.summary()
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'm' for (root).layer_with_weights-0.bias
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.kernel
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer's state 'v' for (root).layer_with_weights-0.bias
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
Model: "sequential_10"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
prune_low_magnitude_dense_10 (None, 20)                822       
_________________________________________________________________
prune_low_magnitude_flatten_ (None, 20)                1         
=================================================================
Total params: 823
Trainable params: 420
Non-trainable params: 403
_________________________________________________________________