Trait burn::tensor::ops::ActivationOps

pub trait ActivationOps<B>
where
    B: Backend,
{
    // Provided methods
    fn leaky_relu(
        tensor: <B as Backend>::FloatTensorPrimitive,
        negative_slope: <B as Backend>::FloatElem,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn relu(
        tensor: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn relu_backward(
        output: <B as Backend>::FloatTensorPrimitive,
        grad: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn gelu(
        tensor: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn prelu(
        tensor: <B as Backend>::FloatTensorPrimitive,
        alpha: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn gelu_backward(
        x: <B as Backend>::FloatTensorPrimitive,
        grad: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn sigmoid(
        tensor: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn sigmoid_backward(
        output: <B as Backend>::FloatTensorPrimitive,
        grad: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn hard_sigmoid(
        tensor: <B as Backend>::FloatTensorPrimitive,
        alpha: <B as Backend>::FloatElem,
        beta: <B as Backend>::FloatElem,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn log_sigmoid(
        tensor: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
    fn log_sigmoid_backward(
        x: <B as Backend>::FloatTensorPrimitive,
        grad: <B as Backend>::FloatTensorPrimitive,
    ) -> <B as Backend>::FloatTensorPrimitive { ... }
}
Expand description

Activation function operations.

This trait let backend implementations override activation functions for better performance.

Provided Methods§

fn leaky_relu( tensor: <B as Backend>::FloatTensorPrimitive, negative_slope: <B as Backend>::FloatElem, ) -> <B as Backend>::FloatTensorPrimitive

Applies the LeakyReLU activation function.

§Arguments
  • tensor - The tensor.
  • negative_slope - The negative_slope value that values smaller than 0 are multiplied with.
§Returns

The output tensor.

fn relu( tensor: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the ReLU activation function.

§Arguments
  • tensor - The tensor.
§Returns

The output tensor.

fn relu_backward( output: <B as Backend>::FloatTensorPrimitive, grad: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the ReLU activation function backward.

§Arguments
  • output - The output tensor.
§Returns

The gradient.

fn gelu( tensor: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the Gelu activation function.

§Arguments
  • tensor - The tensor.
§Returns

The output tensor.

fn prelu( tensor: <B as Backend>::FloatTensorPrimitive, alpha: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the PReLu activation function.

§Arguments
  • tensor - The input tensor
  • alpha - The weight tensor

fn gelu_backward( x: <B as Backend>::FloatTensorPrimitive, grad: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the Gelu activation function backward.

§Arguments
  • x - The tensor.
  • grad - The gradient.
§Returns

The output tensor.

fn sigmoid( tensor: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the Sigmoid activation function.

§Arguments
  • tensor - The tensor.
§Returns

The output tensor.

fn sigmoid_backward( output: <B as Backend>::FloatTensorPrimitive, grad: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the Sigmoid activation function backward.

§Arguments
  • output - The output tensor of the sigmoid function.
  • grad - The gradient.
§Returns

The output tensor.

fn hard_sigmoid( tensor: <B as Backend>::FloatTensorPrimitive, alpha: <B as Backend>::FloatElem, beta: <B as Backend>::FloatElem, ) -> <B as Backend>::FloatTensorPrimitive

Applies the hard Sigmoid activation function.

§Arguments
  • tensor - The tensor.
  • alpha - The alpha value that the tensor is multiplied with.
  • beta - The beta value that is added to the tensor
§Returns

The output tensor.

fn log_sigmoid( tensor: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the LogSigmoid activation function.

§Arguments
  • tensor - The tensor.
§Returns

The output tensor.

fn log_sigmoid_backward( x: <B as Backend>::FloatTensorPrimitive, grad: <B as Backend>::FloatTensorPrimitive, ) -> <B as Backend>::FloatTensorPrimitive

Applies the LogSigmoid activation function backward.

§Arguments
  • x - The input tensor.
  • grad - The gradient.
§Returns

The output gradient.

Object Safety§

This trait is not object safe.

Implementations on Foreign Types§

§

impl<B> ActivationOps<Fusion<B>> for Fusion<B>
where B: FusionBackend,

Implementors§

§

impl<B, C> ActivationOps<Autodiff<B, C>> for Autodiff<B, C>
where B: Backend, C: CheckpointStrategy,

§

impl<E, I, Q> ActivationOps<NdArray<E, I, Q>> for NdArray<E, I, Q>
where E: FloatNdArrayElement, I: IntNdArrayElement, Q: QuantElement,

§

impl<E, Q> ActivationOps<LibTorch<E, Q>> for LibTorch<E, Q>
where E: TchElement, Q: QuantElement,

§

impl<F, I> ActivationOps<Candle<F, I>> for Candle<F, I>
where F: FloatCandleElement, I: IntCandleElement,

§

impl<R, F, I> ActivationOps<JitBackend<R, F, I>> for JitBackend<R, F, I>
where R: JitRuntime, F: FloatElement, I: IntElement,