Skip to content

larq.context

Context managers that configure global behaviour of Larq.

[source]

should_quantize

larq.context.should_quantize()

Returns the current quantized scope.


[source]

quantized_scope

larq.context.quantized_scope(quantize)

A context manager to define the behaviour of QuantizedVariable.

Example

model.save("full_precision_model.h5")  # save full precision latent weights
fp_weights = model.get_weights()  # get latent weights

with larq.context.quantized_scope(True):
    model.save("binary_model.h5")  # save binarized weights
    weights = model.get_weights()  # get binarized weights

Arguments

  • quantize: If should_quantize is True, QuantizedVariable will return their quantized value in the forward pass. If False, QuantizedVariable will act as a latent variable.

[source]

metrics_scope

larq.context.metrics_scope(metrics=[])

A context manager to set the training metrics to be used in quantizers.

Example

with larq.context.metrics_scope(["flip_ratio"]):
    model = tf.keras.models.Sequential(
        [larq.layers.QuantDense(3, kernel_quantizer="ste_sign", input_shape=(32,))]
    )
model.compile(loss="mse", optimizer="sgd")

Arguments

  • metrics: Iterable of metrics to add to quantizers defined inside this context. Currently only the flip_ratio metric is available.

[source]

get_training_metrics

larq.context.get_training_metrics()

Retrieves a live reference to the training metrics in the current scope.

Updating and clearing training metrics using larq.context.metrics_scope is preferred, but get_training_metrics can be used to directly access them.

Example

get_training_metrics().clear()
get_training_metrics().add("flip_ratio")

Returns

A set of training metrics in the current scope.