rand_op

→TensorImplPtr

int rand_op(const int & shape, Dtype dt, Device device, Generator * gen)

Allocates a tensor filled with uniform samples on $[0, 1)$ .

Convenience wrapper around uniform_op with low = 0 / high = 1.

Math

$x_i \sim \mathcal{U}(0, 1)$

See Also

uniform_op — arbitrary $[low, high)$ interval. randn_op — Gaussian variant.

shapeconst Shape&

Shape of the output tensor.

dtDtype

Element dtype. Floating-point dtypes only (F32, F64, BF16, F16).

deviceDevice

Memory domain of the output (CPU or GPU).

genGenerator*= None

Bit-stream state to draw from. When nullptr (the default), uses default_generator.

TensorImplPtr

Newly allocated leaf tensor (requires_grad = false) of the requested shape / dtype / device.