blazefl.reproducibility.RandomStateSnapshot

class blazefl.reproducibility.RandomStateSnapshot(environ: str, python: tuple[Any, ...], numpy: tuple[str, ndarray[tuple[Any, ...], dtype[uint32]], int, int, float], torch_cpu: Tensor, torch_cuda: Tensor | None)

Bases: object

A snapshot of the global random state for major libraries.

This class provides a mechanism to capture the exact state of the global random number generators and later restore it. This is useful for scenarios that require resuming a process from a specific point in time with the identical sequence of random numbers.

environ

The value of the PYTHONHASHSEED environment variable.

Type:

str

python

The internal state of Python’s random module.

Type:

tuple[Any, …]

numpy

The internal state of NumPy’s legacy random number generator.

Type:

tuple[str, numpy.ndarray[tuple[Any, …], numpy.dtype[numpy.uint32]], int, int, float]

torch_cpu

The RNG state of the PyTorch CPU generator.

Type:

torch.Tensor

torch_cuda

The RNG state of the PyTorch CUDA generator, if available.

Type:

torch.Tensor | None

__init__(environ: str, python: tuple[Any, ...], numpy: tuple[str, ndarray[tuple[Any, ...], dtype[uint32]], int, int, float], torch_cpu: Tensor, torch_cuda: Tensor | None) → None

Methods

__init__(environ, python, numpy, torch_cpu, ...)
capture()	Captures the current global random state.
restore(snapshot)	Restores the global random state from a snapshot object.

Attributes

environ
python
numpy
torch_cpu
torch_cuda

classmethod capture() → RandomStateSnapshot

Captures the current global random state.

Returns:

A RandomStateSnapshot instance containing the captured states.

environ: str

numpy: tuple[str, ndarray[tuple[Any, ...], dtype[uint32]], int, int, float]

python: tuple[Any, ...]

static restore(snapshot: RandomStateSnapshot) → None

Restores the global random state from a snapshot object.

Parameters:

snapshot – The RandomStateSnapshot to restore from.

torch_cpu: Tensor

torch_cuda: Tensor | None