hard_swish_

hard_swish_(x: &Tensor<T>, out: &mut Tensor<C> | Tensor<C>) -> Result<Tensor<C>, TensorError>

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

Parameters:

x: Input values out: Tensor to write to

Returns:

Tensor with type C

Examples:

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

fn main() -> Result<(), TensorError> {
    let a = Tensor::<f32>::new([10.0]);
    let b = a.hard_swish_(&mut a.clone())?;
    println!("{}", b);
    assert_eq!(a.ptr().ptr as u64, b.ptr().ptr as u64);
    Ok(())
}

Backend Support

Backend	Supported
CPU	✅
Cuda	✅