Comment on page
nn.linear
fn linear(inputs: Tensor<T>, weights: Tensor<T>, bias: Tensor<T>) -> Tensor<T>
Performs a linear transformation of the input tensor using the provided weights and bias.
tensor(@Tensor<T>) - A 1D tensor representing the input tensor.weights(@Tensor<T>) - A 2D tensor representing the weights.bias(@Tensor<T>) - A 1D tensor representing the bias.
- This function asserts that the input tensor
inputsmust be 1D, weights tensor must be 2D, and bias tensor must be 1D.
A
Tensor<T> representing the result of the linear transformation.use array::{ArrayTrait, SpanTrait};
use orion::operators::tensor::{TensorTrait, Tensor, I32Tensor};
use orion::operators::nn::{NNTrait, I32NN};
use orion::numbers::{i32, IntegerTrait};
fn linear_example() -> Tensor<i32> {
// We instantiate inputs here.
let inputs = TensorTrait::<i32>::new(
shape: array![3].span(),
data: array![
IntegerTrait::new(71, true), IntegerTrait::new(38, false), IntegerTrait::new(62, false),
]
.span(),
);
// We instantiate weights here.
let weights = TensorTrait::<i32>::new(
shape: array![2, 3].span(),
data: array![
IntegerTrait::new(8, true),
IntegerTrait::new(64, false),
IntegerTrait::new(40, false),
IntegerTrait::new(33, true),
IntegerTrait::new(34, true),
IntegerTrait::new(20, true),
]
.span(),
);
// We instantiate bias here.
let bias = TensorTrait::<i32>::new(
shape: array![2].span(),
data: array![IntegerTrait::new(61, false), IntegerTrait::new(61, true),].span(),
);
return NNTrait::linear(inputs, weights, bias);
}
>>> [5541, -250]
Last modified 2mo ago