Panduan komprehensif pelatihan sadar kuantisasi

Lihat di TensorFlow.org Jalankan di Google Colab Lihat sumber di GitHub Unduh buku catatan

Selamat datang di panduan komprehensif untuk pelatihan sadar kuantisasi Keras.

Halaman ini mendokumentasikan berbagai kasus penggunaan dan menunjukkan cara menggunakan API untuk masing-masing kasus. Setelah Anda mengetahui API mana yang Anda butuhkan, temukan parameter dan detail tingkat rendah di dokumen API .

Kasus penggunaan berikut tercakup:

  • Terapkan model dengan kuantisasi 8-bit dengan langkah-langkah ini.
    • Tentukan model sadar kuantisasi.
    • Untuk model Keras HDF5 saja, gunakan checkpointing khusus dan logika deserialisasi. Pelatihan sebaliknya standar.
    • Buat model terkuantisasi dari model sadar kuantisasi.
  • Percobaan dengan kuantisasi.
    • Apa pun untuk eksperimen tidak memiliki jalur yang didukung untuk penerapan.
    • Lapisan Keras khusus termasuk dalam eksperimen.

Mendirikan

Untuk menemukan API yang Anda butuhkan dan memahami tujuan, Anda dapat menjalankan tetapi lewati membaca bagian ini.

! pip uninstall -y tensorflow
! pip install -q tf-nightly
! pip install -q tensorflow-model-optimization

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

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 setup_pretrained_model():
  model = setup_model()
  pretrained_weights = setup_pretrained_weights()
  model.load_weights(pretrained_weights)
  return model

setup_model()
pretrained_weights = setup_pretrained_weights()

Tentukan model sadar kuantisasi

Dengan menentukan model dengan cara berikut, ada jalur yang tersedia untuk penerapan ke backend yang tercantum di halaman ikhtisar . Secara default, kuantisasi 8-bit digunakan.

Kuantisasi seluruh model

Kasus penggunaan Anda:

  • Model subkelas tidak didukung.

Kiat untuk akurasi model yang lebih baik:

  • Coba "Kuantisasi beberapa lapisan" untuk melewati kuantisasi lapisan yang paling mengurangi akurasi.
  • Biasanya lebih baik untuk menyesuaikan dengan pelatihan sadar kuantisasi daripada pelatihan dari awal.

Untuk membuat seluruh model mengetahui kuantisasi, terapkan tfmot.quantization.keras.quantize_model ke model.

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

quant_aware_model = tfmot.quantization.keras.quantize_model(base_model)
quant_aware_model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer (QuantizeLaye (None, 20)                3         
_________________________________________________________________
quant_dense_2 (QuantizeWrapp (None, 20)                425       
_________________________________________________________________
quant_flatten_2 (QuantizeWra (None, 20)                1         
=================================================================
Total params: 429
Trainable params: 420
Non-trainable params: 9
_________________________________________________________________

Kuantisasi beberapa lapisan

Kuantisasi model dapat memiliki efek negatif pada akurasi. Anda dapat secara selektif mengkuantisasi lapisan model untuk mengeksplorasi trade-off antara akurasi, kecepatan, dan ukuran model.

Kasus penggunaan Anda:

  • Untuk menerapkan ke backend yang hanya berfungsi dengan baik dengan model terkuantisasi penuh (misalnya EdgeTPU v1, kebanyakan DSP), coba "Kuantisasi seluruh model".

Kiat untuk akurasi model yang lebih baik:

  • Biasanya lebih baik untuk menyesuaikan dengan pelatihan sadar kuantisasi daripada pelatihan dari awal.
  • Coba kuantisasi lapisan selanjutnya alih-alih lapisan pertama.
  • Hindari mengkuantisasi lapisan kritis (misalnya mekanisme perhatian).

Dalam contoh di bawah ini, kuantisasi hanya lapisan Dense .

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

# Helper function uses `quantize_annotate_layer` to annotate that only the 
# Dense layers should be quantized.
def apply_quantization_to_dense(layer):
  if isinstance(layer, tf.keras.layers.Dense):
    return tfmot.quantization.keras.quantize_annotate_layer(layer)
  return layer

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

# Now that the Dense layers are annotated,
# `quantize_apply` actually makes the model quantization aware.
quant_aware_model = tfmot.quantization.keras.quantize_apply(annotated_model)
quant_aware_model.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 #   
=================================================================
quantize_layer_1 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_3 (QuantizeWrapp (None, 20)                425       
_________________________________________________________________
flatten_3 (Flatten)          (None, 20)                0         
=================================================================
Total params: 428
Trainable params: 420
Non-trainable params: 8
_________________________________________________________________

Sementara contoh ini menggunakan jenis lapisan untuk memutuskan apa yang akan dikuantisasi, cara termudah untuk mengkuantisasi lapisan tertentu adalah dengan mengatur properti name , dan mencari nama itu di clone_function .

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

Lebih mudah dibaca tetapi akurasi model berpotensi lebih rendah

Ini tidak kompatibel dengan finetuning dengan pelatihan sadar kuantisasi, itulah sebabnya mungkin kurang akurat daripada contoh di atas.

Contoh fungsional

# Use `quantize_annotate_layer` to annotate that the `Dense` layer
# should be quantized.
i = tf.keras.Input(shape=(20,))
x = tfmot.quantization.keras.quantize_annotate_layer(tf.keras.layers.Dense(10))(i)
o = tf.keras.layers.Flatten()(x)
annotated_model = tf.keras.Model(inputs=i, outputs=o)

# Use `quantize_apply` to actually make the model quantization aware.
quant_aware_model = tfmot.quantization.keras.quantize_apply(annotated_model)

# For deployment purposes, the tool adds `QuantizeLayer` after `InputLayer` so that the
# quantized model can take in float inputs instead of only uint8.
quant_aware_model.summary()
Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 20)]              0         
_________________________________________________________________
quantize_layer_2 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_4 (QuantizeWrapp (None, 10)                215       
_________________________________________________________________
flatten_4 (Flatten)          (None, 10)                0         
=================================================================
Total params: 218
Trainable params: 210
Non-trainable params: 8
_________________________________________________________________

Contoh berurutan

# Use `quantize_annotate_layer` to annotate that the `Dense` layer
# should be quantized.
annotated_model = tf.keras.Sequential([
  tfmot.quantization.keras.quantize_annotate_layer(tf.keras.layers.Dense(20, input_shape=input_shape)),
  tf.keras.layers.Flatten()
])

# Use `quantize_apply` to actually make the model quantization aware.
quant_aware_model = tfmot.quantization.keras.quantize_apply(annotated_model)

quant_aware_model.summary()
Model: "sequential_4"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_3 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_5 (QuantizeWrapp (None, 20)                425       
_________________________________________________________________
flatten_5 (Flatten)          (None, 20)                0         
=================================================================
Total params: 428
Trainable params: 420
Non-trainable params: 8
_________________________________________________________________

Pos pemeriksaan dan deserialize

Kasus penggunaan Anda: kode ini hanya diperlukan untuk format model HDF5 (bukan bobot HDF5 atau format lainnya).

# Define the model.
base_model = setup_model()
base_model.load_weights(pretrained_weights) # optional but recommended for model accuracy
quant_aware_model = tfmot.quantization.keras.quantize_model(base_model)

# Save or checkpoint the model.
_, keras_model_file = tempfile.mkstemp('.h5')
quant_aware_model.save(keras_model_file)

# `quantize_scope` is needed for deserializing HDF5 models.
with tfmot.quantization.keras.quantize_scope():
  loaded_model = tf.keras.models.load_model(keras_model_file)

loaded_model.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.
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.
WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.
Model: "sequential_5"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_4 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_6 (QuantizeWrapp (None, 20)                425       
_________________________________________________________________
quant_flatten_6 (QuantizeWra (None, 20)                1         
=================================================================
Total params: 429
Trainable params: 420
Non-trainable params: 9
_________________________________________________________________

Buat dan terapkan model terkuantisasi

Secara umum, lihat dokumentasi untuk penerapan backend yang akan Anda gunakan.

Ini adalah contoh untuk backend TFLite.

base_model = setup_pretrained_model()
quant_aware_model = tfmot.quantization.keras.quantize_model(base_model)

# Typically you train the model here.

converter = tf.lite.TFLiteConverter.from_keras_model(quant_aware_model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]

quantized_tflite_model = converter.convert()
1/1 [==============================] - 0s 229ms/step - loss: 16.1181 - accuracy: 0.0000e+00
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.
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.
WARNING:absl:Found untraced functions such as dense_7_layer_call_and_return_conditional_losses, dense_7_layer_call_fn, flatten_7_layer_call_and_return_conditional_losses, flatten_7_layer_call_fn, dense_7_layer_call_fn while saving (showing 5 of 10). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: /tmp/tmph82hcqfi/assets
INFO:tensorflow:Assets written to: /tmp/tmph82hcqfi/assets

Percobaan dengan kuantisasi

Kasus penggunaan Anda : menggunakan API berikut berarti tidak ada jalur yang didukung untuk penerapan. Misalnya, konversi TFLite dan implementasi kernel hanya mendukung kuantisasi 8-bit. Fitur-fiturnya juga eksperimental dan tidak tunduk pada kompatibilitas mundur.

Pengaturan: DefaultDenseQuantizeConfig

Bereksperimen memerlukan penggunaan tfmot.quantization.keras.QuantizeConfig , yang menjelaskan cara mengkuantisasi bobot, aktivasi, dan keluaran sebuah lapisan.

Di bawah ini adalah contoh yang mendefinisikan QuantizeConfig sama yang digunakan untuk lapisan Dense di default API.

Selama propagasi maju dalam contoh ini, LastValueQuantizer dikembalikan dalam get_weights_and_quantizers dipanggil dengan layer.kernel sebagai input, menghasilkan output. Outputnya menggantikan layer.kernel dalam propagasi maju asli dari layer Dense , melalui logika yang didefinisikan dalam set_quantize_weights . Ide yang sama berlaku untuk aktivasi dan output.

LastValueQuantizer = tfmot.quantization.keras.quantizers.LastValueQuantizer
MovingAverageQuantizer = tfmot.quantization.keras.quantizers.MovingAverageQuantizer

class DefaultDenseQuantizeConfig(tfmot.quantization.keras.QuantizeConfig):
    # Configure how to quantize weights.
    def get_weights_and_quantizers(self, layer):
      return [(layer.kernel, LastValueQuantizer(num_bits=8, symmetric=True, narrow_range=False, per_axis=False))]

    # Configure how to quantize activations.
    def get_activations_and_quantizers(self, layer):
      return [(layer.activation, MovingAverageQuantizer(num_bits=8, symmetric=False, narrow_range=False, per_axis=False))]

    def set_quantize_weights(self, layer, quantize_weights):
      # Add this line for each item returned in `get_weights_and_quantizers`
      # , in the same order
      layer.kernel = quantize_weights[0]

    def set_quantize_activations(self, layer, quantize_activations):
      # Add this line for each item returned in `get_activations_and_quantizers`
      # , in the same order.
      layer.activation = quantize_activations[0]

    # Configure how to quantize outputs (may be equivalent to activations).
    def get_output_quantizers(self, layer):
      return []

    def get_config(self):
      return {}

Kuantisasi lapisan Keras khusus

Contoh ini menggunakan DefaultDenseQuantizeConfig untuk mengkuantisasi CustomLayer .

Menerapkan konfigurasi sama di seluruh kasus penggunaan "Eksperimen dengan kuantisasi".

quantize_annotate_layer = tfmot.quantization.keras.quantize_annotate_layer
quantize_annotate_model = tfmot.quantization.keras.quantize_annotate_model
quantize_scope = tfmot.quantization.keras.quantize_scope

class CustomLayer(tf.keras.layers.Dense):
  pass

model = quantize_annotate_model(tf.keras.Sequential([
   quantize_annotate_layer(CustomLayer(20, input_shape=(20,)), DefaultDenseQuantizeConfig()),
   tf.keras.layers.Flatten()
]))

# `quantize_apply` requires mentioning `DefaultDenseQuantizeConfig` with `quantize_scope`
# as well as the custom Keras layer.
with quantize_scope(
  {'DefaultDenseQuantizeConfig': DefaultDenseQuantizeConfig,
   'CustomLayer': CustomLayer}):
  # Use `quantize_apply` to actually make the model quantization aware.
  quant_aware_model = tfmot.quantization.keras.quantize_apply(model)

quant_aware_model.summary()
Model: "sequential_8"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_6 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_custom_layer (Quantize (None, 20)                425       
_________________________________________________________________
quant_flatten_9 (QuantizeWra (None, 20)                1         
=================================================================
Total params: 429
Trainable params: 420
Non-trainable params: 9
_________________________________________________________________

Ubah parameter kuantisasi

Kesalahan umum: mengkuantisasi bias menjadi kurang dari 32-bit biasanya terlalu merusak akurasi model.

Contoh ini memodifikasi layer Dense untuk menggunakan 4-bit untuk bobotnya alih-alih 8-bit default. Model lainnya terus menggunakan default API.

quantize_annotate_layer = tfmot.quantization.keras.quantize_annotate_layer
quantize_annotate_model = tfmot.quantization.keras.quantize_annotate_model
quantize_scope = tfmot.quantization.keras.quantize_scope

class ModifiedDenseQuantizeConfig(DefaultDenseQuantizeConfig):
    # Configure weights to quantize with 4-bit instead of 8-bits.
    def get_weights_and_quantizers(self, layer):
      return [(layer.kernel, LastValueQuantizer(num_bits=4, symmetric=True, narrow_range=False, per_axis=False))]

Menerapkan konfigurasi sama di seluruh kasus penggunaan "Eksperimen dengan kuantisasi".

model = quantize_annotate_model(tf.keras.Sequential([
   # Pass in modified `QuantizeConfig` to modify this Dense layer.
   quantize_annotate_layer(tf.keras.layers.Dense(20, input_shape=(20,)), ModifiedDenseQuantizeConfig()),
   tf.keras.layers.Flatten()
]))

# `quantize_apply` requires mentioning `ModifiedDenseQuantizeConfig` with `quantize_scope`:
with quantize_scope(
  {'ModifiedDenseQuantizeConfig': ModifiedDenseQuantizeConfig}):
  # Use `quantize_apply` to actually make the model quantization aware.
  quant_aware_model = tfmot.quantization.keras.quantize_apply(model)

quant_aware_model.summary()
Model: "sequential_9"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_7 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_9 (QuantizeWrapp (None, 20)                425       
_________________________________________________________________
quant_flatten_10 (QuantizeWr (None, 20)                1         
=================================================================
Total params: 429
Trainable params: 420
Non-trainable params: 9
_________________________________________________________________

Ubah bagian lapisan untuk dikuantisasi

Contoh ini memodifikasi lapisan Dense untuk melewatkan kuantisasi aktivasi. Model lainnya terus menggunakan default API.

quantize_annotate_layer = tfmot.quantization.keras.quantize_annotate_layer
quantize_annotate_model = tfmot.quantization.keras.quantize_annotate_model
quantize_scope = tfmot.quantization.keras.quantize_scope

class ModifiedDenseQuantizeConfig(DefaultDenseQuantizeConfig):
    def get_activations_and_quantizers(self, layer):
      # Skip quantizing activations.
      return []

    def set_quantize_activations(self, layer, quantize_activations):
      # Empty since `get_activaations_and_quantizers` returns
      # an empty list.
      return

Menerapkan konfigurasi sama di seluruh kasus penggunaan "Eksperimen dengan kuantisasi".

model = quantize_annotate_model(tf.keras.Sequential([
   # Pass in modified `QuantizeConfig` to modify this Dense layer.
   quantize_annotate_layer(tf.keras.layers.Dense(20, input_shape=(20,)), ModifiedDenseQuantizeConfig()),
   tf.keras.layers.Flatten()
]))

# `quantize_apply` requires mentioning `ModifiedDenseQuantizeConfig` with `quantize_scope`:
with quantize_scope(
  {'ModifiedDenseQuantizeConfig': ModifiedDenseQuantizeConfig}):
  # Use `quantize_apply` to actually make the model quantization aware.
  quant_aware_model = tfmot.quantization.keras.quantize_apply(model)

quant_aware_model.summary()
Model: "sequential_10"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_8 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_10 (QuantizeWrap (None, 20)                423       
_________________________________________________________________
quant_flatten_11 (QuantizeWr (None, 20)                1         
=================================================================
Total params: 427
Trainable params: 420
Non-trainable params: 7
_________________________________________________________________

Gunakan algoritme kuantisasi khusus

Kelas tfmot.quantization.keras.quantizers.Quantizer adalah callable yang dapat menerapkan algoritme apa pun ke inputnya.

Dalam contoh ini, inputnya adalah bobot, dan kami menerapkan matematika dalam fungsi FixedRangeQuantizer __call__ ke bobot. Alih-alih nilai bobot asli, output dari FixedRangeQuantizer sekarang diteruskan ke apa pun yang akan menggunakan bobot.

quantize_annotate_layer = tfmot.quantization.keras.quantize_annotate_layer
quantize_annotate_model = tfmot.quantization.keras.quantize_annotate_model
quantize_scope = tfmot.quantization.keras.quantize_scope

class FixedRangeQuantizer(tfmot.quantization.keras.quantizers.Quantizer):
  """Quantizer which forces outputs to be between -1 and 1."""

  def build(self, tensor_shape, name, layer):
    # Not needed. No new TensorFlow variables needed.
    return {}

  def __call__(self, inputs, training, weights, **kwargs):
    return tf.keras.backend.clip(inputs, -1.0, 1.0)

  def get_config(self):
    # Not needed. No __init__ parameters to serialize.
    return {}


class ModifiedDenseQuantizeConfig(DefaultDenseQuantizeConfig):
    # Configure weights to quantize with 4-bit instead of 8-bits.
    def get_weights_and_quantizers(self, layer):
      # Use custom algorithm defined in `FixedRangeQuantizer` instead of default Quantizer.
      return [(layer.kernel, FixedRangeQuantizer())]

Menerapkan konfigurasi sama di seluruh kasus penggunaan "Eksperimen dengan kuantisasi".

model = quantize_annotate_model(tf.keras.Sequential([
   # Pass in modified `QuantizeConfig` to modify this `Dense` layer.
   quantize_annotate_layer(tf.keras.layers.Dense(20, input_shape=(20,)), ModifiedDenseQuantizeConfig()),
   tf.keras.layers.Flatten()
]))

# `quantize_apply` requires mentioning `ModifiedDenseQuantizeConfig` with `quantize_scope`:
with quantize_scope(
  {'ModifiedDenseQuantizeConfig': ModifiedDenseQuantizeConfig}):
  # Use `quantize_apply` to actually make the model quantization aware.
  quant_aware_model = tfmot.quantization.keras.quantize_apply(model)

quant_aware_model.summary()
Model: "sequential_11"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
quantize_layer_9 (QuantizeLa (None, 20)                3         
_________________________________________________________________
quant_dense_11 (QuantizeWrap (None, 20)                423       
_________________________________________________________________
quant_flatten_12 (QuantizeWr (None, 20)                1         
=================================================================
Total params: 427
Trainable params: 420
Non-trainable params: 7
_________________________________________________________________