Module candle_nn::linear

Linear layer

This layer applies a linear transformation to the incoming data, y = x@w.t() + b. The bias is optional. The forward method can be used to apply the layer, it supports input with a batch dimension (so of shape (b_sz, in_c)) or without (of shape (in_c,)), the output has shape (b_sz, out_c) and (out_c,) respectively.

use candle::{Tensor, Device::Cpu};
use candle_nn::{Linear, Module};

let w = Tensor::new(&[[1f32, 2.], [3., 4.], [5., 6.]], &Cpu)?;
let layer = Linear::new(w, None); // Use no bias.
let xs = Tensor::new(&[[10f32, 100.]], &Cpu)?;
let ys = layer.forward(&xs)?;
assert_eq!(ys.to_vec2::<f32>()?, &[[210.0, 430.0, 650.0]]);

Structs

• Linear

Functions

• linear
  Create or initialize a new linear layer.
• linear_b
• linear_no_bias
  Create or initialize a new linear layer without biases.