rlightning.utils.utils

General utility functions and classes for RLightning.

This module provides utility functions for device handling, type conversions, and internal flags for controlling runtime behavior.

class rlightning.utils.utils.InternalFlag

Bases: object

Dynamic access to internal runtime flags.

Provides class-level properties to check various runtime modes controlled by environment variables.

Example

>>> if InternalFlag.DEBUG:
...     # do something related with profiling and debugging
...     pass

classmethod get_env_vars() → Dict[str, str]

Get all internal flags as environment variable dictionary.

Returns:

Dictionary mapping environment variable names to values.

rlightning.utils.utils.to_device(data: Any, device: str | MockModule('torch.device')) → Any

Move tensors in data structure to specified device.

Recursively traverses the data structure and moves any Tensor or TensorDict objects to the specified device.

Parameters:

• data – Any data structure that may contain Tensor or TensorDict.

• device – Target device (e.g., ‘cpu’, ‘cuda:0’).

Returns:

Data structure with tensors moved to the specified device.

rlightning.utils.utils.to_numpy(data: Any) → Any

Convert tensors in data structure to NumPy arrays.

Recursively traverses the data structure and converts any Tensor objects to NumPy arrays.

Parameters:

data – Any data structure that may contain Tensor objects.

Returns:

Data structure with tensors converted to NumPy arrays.

rlightning.utils.utils.torch_dtype_from_precision(precision: int | str | None) → MockModule('torch.dtype') | None

Convert precision specification to torch dtype.

Parameters:

precision – Precision specification (int, str, or None).

Returns:

Corresponding torch dtype, or None if precision is None.

Raises:

ValueError – If precision cannot be parsed to a valid dtype.