tfmot.quantization.keras.quantize_annotate_layer | TensorFlow Model Optimization (original) (raw)
tfmot.quantization.keras.quantize_annotate_layer
Stay organized with collections Save and categorize content based on your preferences.
Annotate a tf.keras layer to be quantized.
tfmot.quantization.keras.quantize_annotate_layer(
to_annotate, quantize_config=None
)
Used in the notebooks
| Used in the guide |
|---|
| Quantization aware training comprehensive guide |
This function does not actually quantize the layer. It is merely used to specify that the layer should be quantized. The layer then gets quantized accordingly when quantize_apply is used.
This method should be used when the user wants to quantize only certain layers of the model, or change the default behavior of how a layer is quantized.
Annotate a layer:
model = keras.Sequential([
layers.Dense(10, activation='relu', input_shape=(100,)),
quantize_annotate_layer(layers.Dense(2, activation='sigmoid'))
])
# Only the second Dense layer is quantized.
quantized_model = quantize_apply(model)
| Args | |
|---|---|
| to_annotate | tf.keras layer which needs to be quantized. |
| quantize_config | optional QuantizeConfig which controls how the layer is quantized. In its absence, the default behavior for the layer is used. |
| Returns |
|---|
| tf.keras layer wrapped with QuantizeAnnotate. |
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2023-05-26 UTC.