Trait burn_tensor::ops::ModuleOps

pub trait ModuleOps<B: Backend> {
    // Required methods
    fn conv2d(
        x: FloatTensor<B, 4>,
        weight: FloatTensor<B, 4>,
        bias: Option<FloatTensor<B, 1>>,
        options: ConvOptions<2>
    ) -> FloatTensor<B, 4>;
    fn conv_transpose2d(
        x: FloatTensor<B, 4>,
        weight: FloatTensor<B, 4>,
        bias: Option<FloatTensor<B, 1>>,
        options: ConvTransposeOptions<2>
    ) -> FloatTensor<B, 4>;
    fn avg_pool2d(
        x: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        stride: [usize; 2],
        padding: [usize; 2],
        count_include_pad: bool
    ) -> FloatTensor<B, 4>;
    fn avg_pool2d_backward(
        x: FloatTensor<B, 4>,
        grad: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        stride: [usize; 2],
        padding: [usize; 2],
        count_include_pad: bool
    ) -> FloatTensor<B, 4>;
    fn adaptive_avg_pool2d(
        x: FloatTensor<B, 4>,
        output_size: [usize; 2]
    ) -> FloatTensor<B, 4>;
    fn adaptive_avg_pool2d_backward(
        x: FloatTensor<B, 4>,
        grad: FloatTensor<B, 4>
    ) -> FloatTensor<B, 4>;
    fn max_pool2d(
        x: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        stride: [usize; 2],
        padding: [usize; 2],
        dilation: [usize; 2]
    ) -> FloatTensor<B, 4>;
    fn max_pool2d_with_indices(
        x: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        stride: [usize; 2],
        padding: [usize; 2],
        dilation: [usize; 2]
    ) -> MaxPool2dWithIndices<B>;
    fn max_pool2d_with_indices_backward(
        x: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        stride: [usize; 2],
        padding: [usize; 2],
        dilation: [usize; 2],
        output_grad: FloatTensor<B, 4>,
        indices: IntTensor<B, 4>
    ) -> MaxPool2dBackward<B>;
    fn interpolate(
        x: FloatTensor<B, 4>,
        output_size: [usize; 2],
        options: InterpolateOptions
    ) -> FloatTensor<B, 4>;
    fn interpolate_backward(
        x: FloatTensor<B, 4>,
        grad: FloatTensor<B, 4>,
        output_size: [usize; 2],
        options: InterpolateOptions
    ) -> FloatTensor<B, 4>;

    // Provided methods
    fn embedding(
        weights: FloatTensor<B, 2>,
        indices: IntTensor<B, 2>
    ) -> FloatTensor<B, 3> { ... }
    fn embedding_backward(
        weights: FloatTensor<B, 2>,
        output_grad: FloatTensor<B, 3>,
        indices: IntTensor<B, 2>
    ) -> FloatTensor<B, 2> { ... }
    fn conv1d(
        x: FloatTensor<B, 3>,
        weight: FloatTensor<B, 3>,
        bias: Option<FloatTensor<B, 1>>,
        options: ConvOptions<1>
    ) -> FloatTensor<B, 3> { ... }
    fn conv1d_backward(
        x: FloatTensor<B, 3>,
        weight: FloatTensor<B, 3>,
        bias: Option<FloatTensor<B, 1>>,
        output_grad: FloatTensor<B, 3>,
        options: ConvOptions<1>
    ) -> Conv1dBackward<B> { ... }
    fn conv2d_backward(
        x: FloatTensor<B, 4>,
        weight: FloatTensor<B, 4>,
        bias: Option<FloatTensor<B, 1>>,
        output_grad: FloatTensor<B, 4>,
        options: ConvOptions<2>
    ) -> Conv2dBackward<B> { ... }
    fn conv_transpose1d(
        x: FloatTensor<B, 3>,
        weight: FloatTensor<B, 3>,
        bias: Option<FloatTensor<B, 1>>,
        options: ConvTransposeOptions<1>
    ) -> FloatTensor<B, 3> { ... }
    fn conv_transpose1d_backward(
        x: FloatTensor<B, 3>,
        weight: FloatTensor<B, 3>,
        bias: Option<FloatTensor<B, 1>>,
        output_grad: FloatTensor<B, 3>,
        options: ConvTransposeOptions<1>
    ) -> Conv1dBackward<B> { ... }
    fn conv_transpose2d_backward(
        x: FloatTensor<B, 4>,
        weight: FloatTensor<B, 4>,
        bias: Option<FloatTensor<B, 1>>,
        output_grad: FloatTensor<B, 4>,
        options: ConvTransposeOptions<2>
    ) -> Conv2dBackward<B> { ... }
    fn unfold4d(
        x: FloatTensor<B, 4>,
        kernel_size: [usize; 2],
        options: UnfoldOptions
    ) -> FloatTensor<B, 3> { ... }
    fn avg_pool1d(
        x: FloatTensor<B, 3>,
        kernel_size: usize,
        stride: usize,
        padding: usize,
        count_include_pad: bool
    ) -> FloatTensor<B, 3> { ... }
    fn avg_pool1d_backward(
        x: FloatTensor<B, 3>,
        grad: FloatTensor<B, 3>,
        kernel_size: usize,
        stride: usize,
        padding: usize,
        count_include_pad: bool
    ) -> FloatTensor<B, 3> { ... }
    fn adaptive_avg_pool1d(
        x: FloatTensor<B, 3>,
        output_size: usize
    ) -> FloatTensor<B, 3> { ... }
    fn adaptive_avg_pool1d_backward(
        x: FloatTensor<B, 3>,
        grad: FloatTensor<B, 3>
    ) -> FloatTensor<B, 3> { ... }
    fn max_pool1d(
        x: FloatTensor<B, 3>,
        kernel_size: usize,
        stride: usize,
        padding: usize,
        dilation: usize
    ) -> FloatTensor<B, 3> { ... }
    fn max_pool1d_with_indices(
        x: FloatTensor<B, 3>,
        kernel_size: usize,
        stride: usize,
        padding: usize,
        dilation: usize
    ) -> MaxPool1dWithIndices<B> { ... }
    fn max_pool1d_with_indices_backward(
        x: FloatTensor<B, 3>,
        kernel_size: usize,
        stride: usize,
        padding: usize,
        dilation: usize,
        output_grad: FloatTensor<B, 3>,
        indices: IntTensor<B, 3>
    ) -> MaxPool1dBackward<B> { ... }
}

Module operations trait.

Required Methods

fn conv2d( x: FloatTensor<B, 4>, weight: FloatTensor<B, 4>, bias: Option<FloatTensor<B, 1>>, options: ConvOptions<2> ) -> FloatTensor<B, 4>

Two dimensional convolution.

Shapes

x: [batch_size, channels_in, height, width], weight: [channels_out, channels_in, kernel_size_1, kernel_size_2], bias: [channels_out],

fn conv_transpose2d( x: FloatTensor<B, 4>, weight: FloatTensor<B, 4>, bias: Option<FloatTensor<B, 1>>, options: ConvTransposeOptions<2> ) -> FloatTensor<B, 4>

Two dimensional transposed convolution.

Shapes

x: [batch_size, channels_in, height, width], weight: [channels_in, channels_out, kernel_size_1, kernel_size_2], bias: [channels_out],

fn avg_pool2d( x: FloatTensor<B, 4>, kernel_size: [usize; 2], stride: [usize; 2], padding: [usize; 2], count_include_pad: bool ) -> FloatTensor<B, 4>

Two dimensional avg pooling.

Shapes

x: [batch_size, channels, height, width],

fn avg_pool2d_backward( x: FloatTensor<B, 4>, grad: FloatTensor<B, 4>, kernel_size: [usize; 2], stride: [usize; 2], padding: [usize; 2], count_include_pad: bool ) -> FloatTensor<B, 4>

Backward pass for the avg pooling 2d operation.

fn adaptive_avg_pool2d( x: FloatTensor<B, 4>, output_size: [usize; 2] ) -> FloatTensor<B, 4>

Two dimensional adaptive avg pooling.

Shapes

x: [batch_size, channels, height, width],

fn adaptive_avg_pool2d_backward( x: FloatTensor<B, 4>, grad: FloatTensor<B, 4> ) -> FloatTensor<B, 4>

Backward pass for the adaptive avg pooling 2d operation.

fn max_pool2d( x: FloatTensor<B, 4>, kernel_size: [usize; 2], stride: [usize; 2], padding: [usize; 2], dilation: [usize; 2] ) -> FloatTensor<B, 4>

Two dimensional max pooling.

Shapes

x: [batch_size, channels, height, width],

fn max_pool2d_with_indices( x: FloatTensor<B, 4>, kernel_size: [usize; 2], stride: [usize; 2], padding: [usize; 2], dilation: [usize; 2] ) -> MaxPool2dWithIndices<B>

Two dimensional max pooling with indices.

Shapes

x: [batch_size, channels, height, width],

fn max_pool2d_with_indices_backward( x: FloatTensor<B, 4>, kernel_size: [usize; 2], stride: [usize; 2], padding: [usize; 2], dilation: [usize; 2], output_grad: FloatTensor<B, 4>, indices: IntTensor<B, 4> ) -> MaxPool2dBackward<B>

Backward pass for the max pooling 2d operation.

fn interpolate( x: FloatTensor<B, 4>, output_size: [usize; 2], options: InterpolateOptions ) -> FloatTensor<B, 4>

Down/up samples the input.

Shapes

x: [batch_size, channels, height, width],

fn interpolate_backward( x: FloatTensor<B, 4>, grad: FloatTensor<B, 4>, output_size: [usize; 2], options: InterpolateOptions ) -> FloatTensor<B, 4>

Backward pass for the interpolate operation.

Provided Methods

fn embedding( weights: FloatTensor<B, 2>, indices: IntTensor<B, 2> ) -> FloatTensor<B, 3>

Embedding operation.

Arguments

• weights - The embedding weights.
• indices - The indices tensor.

Returns

The output tensor.

fn embedding_backward( weights: FloatTensor<B, 2>, output_grad: FloatTensor<B, 3>, indices: IntTensor<B, 2> ) -> FloatTensor<B, 2>

Embedding backward operation.

Arguments

• weights - The embedding weights.
• output_grad - The output gradient.
• indices - The indices tensor.

Returns

The gradient.

fn conv1d( x: FloatTensor<B, 3>, weight: FloatTensor<B, 3>, bias: Option<FloatTensor<B, 1>>, options: ConvOptions<1> ) -> FloatTensor<B, 3>

One dimensional convolution.

Shapes

x: [batch_size, channels_in, length], weight: [channels_out, channels_in, kernel_size], bias: [channels_out],

fn conv1d_backward( x: FloatTensor<B, 3>, weight: FloatTensor<B, 3>, bias: Option<FloatTensor<B, 1>>, output_grad: FloatTensor<B, 3>, options: ConvOptions<1> ) -> Conv1dBackward<B>

Backward pass for the conv1d operation.

fn conv2d_backward( x: FloatTensor<B, 4>, weight: FloatTensor<B, 4>, bias: Option<FloatTensor<B, 1>>, output_grad: FloatTensor<B, 4>, options: ConvOptions<2> ) -> Conv2dBackward<B>

Backward pass for the conv2d operation.

fn conv_transpose1d( x: FloatTensor<B, 3>, weight: FloatTensor<B, 3>, bias: Option<FloatTensor<B, 1>>, options: ConvTransposeOptions<1> ) -> FloatTensor<B, 3>

One dimensional transposed convolution.

Shapes

x: [batch_size, channels_in, length], weight: [channels_in, channels_out, length], bias: [channels_out],

fn conv_transpose1d_backward( x: FloatTensor<B, 3>, weight: FloatTensor<B, 3>, bias: Option<FloatTensor<B, 1>>, output_grad: FloatTensor<B, 3>, options: ConvTransposeOptions<1> ) -> Conv1dBackward<B>

Backward pass for the conv transpose 1d operation.

fn conv_transpose2d_backward( x: FloatTensor<B, 4>, weight: FloatTensor<B, 4>, bias: Option<FloatTensor<B, 1>>, output_grad: FloatTensor<B, 4>, options: ConvTransposeOptions<2> ) -> Conv2dBackward<B>

Backward pass for the conv transpose 2d operation.

fn unfold4d( x: FloatTensor<B, 4>, kernel_size: [usize; 2], options: UnfoldOptions ) -> FloatTensor<B, 3>

Four-dimensional unfolding.

Shapes

x: [batch_size, channels_in, height, width], returns: [batch_size, channels_in * kernel_size_1 * kernel_size_2, number of blocks],

fn avg_pool1d( x: FloatTensor<B, 3>, kernel_size: usize, stride: usize, padding: usize, count_include_pad: bool ) -> FloatTensor<B, 3>

One dimensional avg pooling.

Shapes

x: [batch_size, channels, length],

fn avg_pool1d_backward( x: FloatTensor<B, 3>, grad: FloatTensor<B, 3>, kernel_size: usize, stride: usize, padding: usize, count_include_pad: bool ) -> FloatTensor<B, 3>

Backward pass for the avg pooling 1d operation.

fn adaptive_avg_pool1d( x: FloatTensor<B, 3>, output_size: usize ) -> FloatTensor<B, 3>

One dimensional adaptive avg pooling.

Shapes

x: [batch_size, channels, length],

fn adaptive_avg_pool1d_backward( x: FloatTensor<B, 3>, grad: FloatTensor<B, 3> ) -> FloatTensor<B, 3>

Backward pass for the adaptive avg pooling 1d operation.

fn max_pool1d( x: FloatTensor<B, 3>, kernel_size: usize, stride: usize, padding: usize, dilation: usize ) -> FloatTensor<B, 3>

One dimensional max pooling.

Shapes

x: [batch_size, channels, length],

fn max_pool1d_with_indices( x: FloatTensor<B, 3>, kernel_size: usize, stride: usize, padding: usize, dilation: usize ) -> MaxPool1dWithIndices<B>

One dimensional max pooling with indices.

Shapes

x: [batch_size, channels, height, width],

fn max_pool1d_with_indices_backward( x: FloatTensor<B, 3>, kernel_size: usize, stride: usize, padding: usize, dilation: usize, output_grad: FloatTensor<B, 3>, indices: IntTensor<B, 3> ) -> MaxPool1dBackward<B>

Backward pass for the max pooling 1d operation.

Object Safety

This trait is not object safe.

Implementors