module

creation

32 members

lucid.creation

Tensor construction, and the seeding of the default generator.

Thirty entry points: conversion (tensor, as_tensor, from_numpy), filled and *_like constructors, the range builders arange / linspace, the sampling family, the generator controls (seed, get_rng_state / set_rng_state, initial_seed, randperm) and DLPack exchange.

Every constructor takes device=, and passing it is what keeps a pipeline on one device: a helper built without it lands on the CPU and forces a round trip the moment it meets Metal data.

CPU and Metal draw from separate generators, so one seed does not reproduce one stream across both — comparing devices means comparing deterministic ops, or eval() for anything stochastic.

Classes

classGenerator6 methods

A Philox-4×32 random stream of its own.

Functions

fntensor→ Tensor

Construct a new Tensor from Python data, a NumPy array, or another Tensor.

fnas_tensor→ Tensor

Convert data to a tensor, avoiding a copy when the source already matches.

fnfrom_numpy→ Tensor

Copy a NumPy ndarray into an owned Lucid tensor.

fnzeros→ Tensor

Return a tensor filled with the additive identity element, zero.

fnones→ Tensor

Return a tensor filled with the multiplicative identity element, one.

fnempty→ Tensor

Return a tensor whose elements are uninitialised (undefined memory).

fnempty_strided→ Tensor

Create an uninitialised tensor with the specified size and stride.

fnfull→ Tensor

Return a tensor filled with a constant scalar value.

fneye→ Tensor

Return a 2-D matrix with ones on the main diagonal and zeros elsewhere.

fnarange→ Tensor

Return a 1-D tensor of evenly spaced values over a half-open interval.

fnlinspace→ Tensor

Return a 1-D tensor of steps equally spaced values over a closed interval.

fnzeros_like→ Tensor

Return a zero-filled tensor with the same shape, dtype, and device as input.

fnones_like→ Tensor

Return an all-ones tensor with the same shape, dtype, and device as input.

fnempty_like→ Tensor

Return an uninitialised tensor with the same shape, dtype, and device as input.

fnfull_like→ Tensor

Return a constant-filled tensor with the same shape, dtype, and device as input.

fnlogspace→ Tensor

Return a 1-D tensor of steps values evenly spaced on a logarithmic scale.

fnmanual_seed→ Generator

Set the key to seed and rewind the stream to its start.

fnrand→ Tensor

Return a tensor of samples drawn from the continuous uniform distribution.

fnrandn→ Tensor

Return a tensor of samples drawn from the standard normal distribution.

fnrandint→ Tensor

Return a tensor of samples drawn from the discrete uniform distribution.

fnbernoulli→ Tensor

Return a tensor of independent Bernoulli trials with success probability input.

fnnormal→ Tensor

Return a tensor of samples drawn from a parametric normal distribution.

fnrand_like→ Tensor

Return a $U[0,1)$ random tensor with the same shape, dtype, and device as input.

fnrandn_like→ Tensor

Return a $\mathcal{N}(0,1)$ random tensor with the same shape, dtype, and device as input.

fnfrom_dlpack→ Tensor

Construct a Lucid tensor from any object exposing the DLPack protocol (__dlpack__) or from a raw PyCapsule.

fnget_rng_state→ Tensor

Snapshot the full state of the default random generator.

fninitial_seed→ int

Return the seed (key) of the default generator as it was last set.

fnrandperm→ Tensor

Return a uniformly random permutation of the integers $0, 1, \ldots, n-1$ .

fnseed→ int

Seed the default generator from the OS cryptographic entropy source.

fnset_rng_state→ None

Restore the default random generator to a previously captured state.

fnto_dlpack→ object

Export tensor as a DLPack PyCapsule.