ReLU — PyTorch 2.7 documentation (original) (raw)

class torch.nn.ReLU(inplace=False)[source][source]¶

Applies the rectified linear unit function element-wise.

ReLU(x)=(x)+=max⁡(0,x)\text{ReLU}(x) = (x)^+ = \max(0, x)

Parameters

inplace (bool) – can optionally do the operation in-place. Default: False

Shape:

• Input: (∗)(*), where ∗* means any number of dimensions.
• Output: (∗)(*), same shape as the input.

Examples:

m = nn.ReLU() input = torch.randn(2) output = m(input)

An implementation of CReLU - https://arxiv.org/abs/1603.05201

m = nn.ReLU() input = torch.randn(2).unsqueeze(0) output = torch.cat((m(input), m(-input)))