bitorch_engine.layers.qconv.binary.cpp.layer.BinaryConv2dForward

class bitorch_engine.layers.qconv.binary.cpp.layer.BinaryConv2dForward(*args, **kwargs)[source]

A custom autograd function to perform forward pass of a 2D binary convolution.

This class implements a static method forward to carry out the convolution operation using binary weights and activations. The operation is performed using a custom C++ backend for efficiency.

- No class-level attributes.
- forward

Performs the forward pass of the binary convolution.

Methods

forward

Forward pass for the 2D binary convolution.

Attributes

static forward(ctx, activations: Tensor, weights: Tensor, m: int, n: int, k: int, kernel_size: int, stride: int, padding: int, dilation: int, output_edge: int) → Tensor[source]

Forward pass for the 2D binary convolution.

Utilizes a C++ backend implemented in binary_conv_cpp.forward to perform the operation. This method is statically defined and automatically integrated with PyTorch’s autograd mechanism.

Parameters:

  • ctx (-) – Context object that can be used to stash information for backward computation. You can cache arbitrary objects for use in the backward pass using the save_for_backward method.

  • activations (-) – The input feature map or activation tensor.

  • weights (-) – The binary weights tensor.

  • m (-) – Dimensions of the input, specifically: - m: The number of output channels. - n: The number of input channels. - k: The spatial size of the output feature map.

  • n (int) – Dimensions of the input, specifically: - m: The number of output channels. - n: The number of input channels. - k: The spatial size of the output feature map.

  • k (int) – Dimensions of the input, specifically: - m: The number of output channels. - n: The number of input channels. - k: The spatial size of the output feature map.

  • kernel_size (-) – Size of the conv kernel.

  • stride (-) – Stride of the convolution.

  • padding (-) – Zero-padding added to both sides of the input.

  • dilation (-) – Spacing between kernel elements.

  • output_edge (-) – The size of the output edge to ensure the output dimension matches expectations.

Returns:

The output feature map resulting from the binary convolution operation.

Return type:

  • Tensor

Note

This method is part of the forward pass and needs to be paired with a corresponding backward method to enable gradient computation.