bitorch.quantizations.dorefa.InputDoReFa

class bitorch.quantizations.dorefa.InputDoReFa(bits: Optional[int] = None)

Module for applying the dorefa function on inputs.

Reference: “DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients” Zouh et al. 2016, https://arxiv.org/abs/1606.06160

Methods

__init__	Initiates quantization bits.
quantize	DoReFas the tensor to desired bit resolution.

Attributes

bit_width
name

__init__(bits: Optional[int] = None) → None

Initiates quantization bits.

Parameters:

bits (int, optional) – number of bits to quantize into. Defaults to None.

quantize(x: Tensor) → Tensor

DoReFas the tensor to desired bit resolution.

Parameters:

x (torch.Tensor) – tensor to be forwarded.

Returns:

DoReFaed tensor x

Return type:

torch.Tensor