overrides

Attributes

fastforward.overrides.override = forward_override module-attribute

Classes

fastforward.overrides.DisableQuantizationOverride()

Override to disable quantization.

Attach DisableQuantizationOverride instance as quantizer override to disable quantization. Quantization is enabled/disabled using the enable_quantization and disable_quantization methods for all quantizers the instance is attached to.

Quantizers can be 'bulk' attached to using the attach_to and detach methods. Note that detach will only detach from quantizers which where attached to using the attach_to methods, i.e., if an instance of DisableQuantizationOverride is registered as override to a quantizers using different means, it must be detached separately.

[source]

Attributes

quantization_enabled property

True if quantization is enabled, False otherwise.

Methods

attach_to(quantizers)

Attach this override to one or more quantizers.

Parameters:

Name	Type	Description	Default
quantizers	Quantizer | QuantizerCollection | Iterable[Quantizer]	Either a single quantizer, a QuantizerCollection obtained using ff.find_quantizers or an iterable of Quantizers. Attach this override to each.	required

[source]

detach()

Detach this override.

Detach from all quantizers it was attached to using the attach_to method.

[source]

disable_quantization()

Disable quantization.

See the docstring of enable_quantization for more information.

[source]

enable_quantization(enabled=True)

Enable quantization.

More specifically, this instance will not disable quantization for any quantizers it is attached to. Other instance, or other methods may still disable quantization.

Parameters:

Name	Type	Description	Default
enabled	bool	True if quantization must be enabled, False if it must be disabled.	True

[source]

Functions

fastforward.overrides.disable_quantization(model)

Disable quantization for all quantizers in model within context.

The global strict_quantization flag is also set to False during the context.

Parameters:

Name	Type	Description	Default
model	Module	The model for which all quantizers are disabled.	required

[source]

fastforward.overrides.enable_quantization(model)

Enable quantization for all quantizers in model within context.

Note that this context manager does not change the strict_quantization flag. To also (temporarily) change the strict_quantization flag use fastforward.quantization.strict_quantization.strict_quantization_for_module

Parameters:

Name	Type	Description	Default
model	Module	The model for which all quantizers are enabled.	required

[source]