converter.Quantize_Network

converter.Quantize_Network(self, w_alpha, dynamic_alpha=False)

A class to perform quantization on a neural network.

Parameters

Name	Type	Description	Default
w_alpha	float	The alpha value for the quantization. Default is 1.	required
dynamic_alpha	bool	Whether to use dynamic alpha for quantization. Default is False.	False

Attributes

Name	Type	Description
w_alpha	float	The alpha value for the quantization.
dynamic_alpha	bool	Whether to use dynamic alpha for quantization.
v_threshold	float or None	The threshold for the quantization. Default is None.
w_bits	int	The number of bits to use for the quantization.
w_delta	float	The delta value for the quantization.
weight_quant	weight_quantize_fn	The weight quantization function.

Examples

>>> q_net = Quantize_Network(w_alpha=1, dynamic_alpha=True)
>>> q_net.quantize(some_model)

Methods

Name	Description
quantize	Performs quantization on a model.
quantize_block	Performs quantization on a block of a model.

quantize

converter.Quantize_Network.quantize(self, model)

Performs quantization on a model.

Parameters

Name	Type	Description	Default
model	torch.nn.Module	The input model.	required

Returns

Type	Description
torch.nn.Module	The quantized model.

Examples

>>> q_net = Quantize_Network(w_alpha=1, dynamic_alpha=True)
>>> q_net.quantize(some_model)

quantize_block

converter.Quantize_Network.quantize_block(self, model)

Performs quantization on a block of a model.

Parameters

Name	Type	Description	Default
model	torch.nn.Module	The input model.	required

Returns

Type	Description
torch.nn.Module	The quantized model.

Examples

>>> q_net = Quantize_Network(w_alpha=1, dynamic_alpha=True)
>>> q_net.quantize_block(some_model)