Struct TestInputBuilder 

pub struct TestInputBuilder { /* private fields */ }

Fluent builder for TestInput.

Use TestInput::builder to start one. The builder holds the shape, dtype, and stride spec. Call a finalizer (arange, eye, zeros, uniform, bernoulli, random, custom) to produce a TestInput ready to generate a tensor handle, host data, or test tensor.

Example

use cubek_test_utils::{TestInput, StrideSpec, Distribution};

let (handle, host) = TestInput::builder(client, [4, 4])
    .stride(StrideSpec::ColMajor)
    .uniform( 0, -1.0, 1.0)
    .generate_with_f32_host_data();

Implementations

impl TestInputBuilder

pub fn dtype(self, dtype: impl Into<InputDataType>) -> Self

Override the dtype. Defaults to f32.

pub fn stride(self, stride_spec: StrideSpec) -> Self

Override the stride layout. Defaults to StrideSpec::RowMajor.

pub fn arange(self) -> TestInput

0, 1, 2, … in row-major order.

pub fn arange_scaled(self, scale: f32) -> TestInput

arange with each value multiplied by scale.

pub fn eye(self) -> TestInput

Identity matrix (1 on the diagonal, 0 elsewhere).

pub fn zeros(self) -> TestInput

All-zeros tensor.

pub fn random(self, seed: u64, distribution: Distribution) -> TestInput

Random tensor with a custom Distribution.

pub fn uniform(self, seed: u64, lo: f32, hi: f32) -> TestInput

Uniform random in [lo, hi].

pub fn bernoulli(self, seed: u64, p: f32) -> TestInput

Bernoulli random with probability p of 1.

pub fn normal(self, seed: u64, mean: f32, std: f32) -> TestInput

Normal (Gaussian) random with the given mean and std.

pub fn custom(self, data: Vec<f32>) -> TestInput

Tensor populated from an explicit row-major Vec<f32>.

pub fn linspace(self, start: f32, end: f32) -> TestInput

Evenly-spaced values from start to end inclusive, populated in row-major order. The number of points equals the tensor’s element count.

Equivalent to NumPy’s np.linspace(start, end, num=shape.numel()).reshape(shape).