burn_ndarray/ops/
activations.rs

1use crate::{
2    NdArray, NdArrayTensor, SharedArray,
3    element::{FloatNdArrayElement, IntNdArrayElement, QuantElement},
4    execute_with_numeric_dtype,
5    ops::NdArrayMathOps,
6};
7use burn_backend::{ElementConversion, TensorMetadata, ops::ActivationOps, tensor::FloatTensor};
8
9impl<E: FloatNdArrayElement, I: IntNdArrayElement, Q: QuantElement> ActivationOps<Self>
10    for NdArray<E, I, Q>
11where
12    NdArrayTensor: From<SharedArray<E>>,
13    NdArrayTensor: From<SharedArray<I>>,
14{
15    fn relu(tensor: FloatTensor<Self>) -> FloatTensor<Self> {
16        execute_with_numeric_dtype!(tensor, |array| NdArrayMathOps::clamp_min(array, 0.elem()))
17    }
18}