hard_swish

hard_swish(x: &Tensor<T>) -> Result<Tensor<C>, TensorError>

Compute $\large x \cdot \text{min}(\text{max}(0, \frac{x}{6} + 0.5), 1)$ for all elements. A piece-wise linear approximation of the swish function.

Parameters:

x: Input values

Returns:

Tensor with type C

Examples:

use hpt::{ops::FloatUnaryOps, Tensor, error::TensorError};

fn main() -> Result<(), TensorError> {
    let a = Tensor::<f32>::new([2.0]);
    let b = a.hard_swish()?;
    println!("{}", b);  // prints: 1.6666666
    Ok(())
}

Backend Support

Backend	Supported
CPU	✅
Cuda	✅