lucid.save

lucid.save(obj: Tensor | Module | OrderedDict, path: Path | str) → Path

The save function serializes and saves a Lucid object to disk. It supports saving Tensor objects and raw OrderedDict-based state_dicts. The data is stored using Python’s pickle module.

Function Signature

def save(obj: Tensor | Module | OrderedDict, path: Path | str) -> Path

Parameters

  • obj (Tensor | Module | OrderedDict): The object to be saved. Must be one of the following:

    • Tensor: saved as a NumPy array into a .lct file.

    • Module: converted into a state_dict and saved as .lcd.

    • OrderedDict: treated as a raw state_dict and saved as .lcd.

  • path (Path | str): File path to save the object to. If no suffix is provided, an appropriate one will be inferred based on the object type.

Returns

  • Path: The absolute path to the saved file.

Behavior

  • If the suffix is missing:

    • .lct is used for Tensor

    • .lcd is used for OrderedDict or Module

  • For Tensor, only its raw data (as NumPy array) is saved. Autograd-related properties are discarded.

Warning

Tensor objects are saved without gradients, computation graphs, or hooks.

Examples

import lucid
from lucid import save
from pathlib import Path

t = lucid.Tensor([[1.0, 2.0]], requires_grad=True)
path = save(t, "tensor")
print(path.name)  # tensor.lct

model = lucid.nn.Linear(2, 3)
path = save(model.state_dict(), "weights")
print(path.name)  # weights.lcd

Note

This function uses pickle for serialization. The saved files are not guaranteed to be portable across major Python versions or environments.