eazygrad.nn.Linear

class eazygrad.nn.Linear(n_in: int, n_out: int, bias: bool = True, requires_grad: bool = True)

Bases: Module

Fully connected linear layer.

Parameters:

• n_in (int) – Number of input features.

• n_out (int) – Number of output features.

• bias (bool, default=True) – Whether to include a learnable bias term.

• requires_grad (bool, default=True) – Whether the layer parameters should participate in automatic differentiation.

Notes

The layer stores weights with shape (n_in, n_out) and applies the transformation x @ weights + bias to batched 2D inputs.

forward(x: _Tensor) → _Tensor

Apply the forward pass of the module to the input x