deserialize¶

Reconstruct a Tensor from a dict previously produced by serialize.

deserialize ¶

deserialize(data: dict, device: Union[str, device] = 'cpu') -> Tensor

Reconstruct a Tensor from a dict previously produced by serialize.

Parameters:

Name	Type	Description	Default
`data`	`dict`	Dict previously produced by `serialize`.	required
`device`	`str or device`	Device to place all tensors (data blocks and intertwiner weights) on. Defaults to `"cpu"`.	`'cpu'`

Returns:

Type	Description
`Tensor`	Reconstructed Tensor with all blocks on device.

Examples:

>>> payload = torch.load("tensor.tnsr", weights_only=True)
>>> t = deserialize(payload, device="cpu")

Description¶

deserialize is the inverse of serialize. It reads the plain dict produced by serialize (or loaded from disk via torch.load(..., weights_only=True)) and reconstructs the original Tensor, including all indices, block data, and — for non-Abelian tensors — intertwiner weights.

All torch.Tensor values in the result are placed on device.

Parameters¶

Parameter	Type	Default	Description
`data`	`dict`	—	Dict previously produced by `serialize`.
`device`	`str` or `torch.device`	`"cpu"`	Device to place all block data and intertwiner weights on.

Returns¶

Tensor — Fully reconstructed tensor with all blocks on device.

Raises¶

ValueError — If version from data["version"] is not supported.

Notes¶

device applies to both block data tensors ("data") and intertwiner weight tensors ("intw"), if present.
For non-Abelian tensors, reconstruction of the CG data is delegated to nicole.symmetry.delegate.deserialize.
The dtype is stored in the serialized dict and applied to all tensors on reconstruction; no explicit dtype argument is needed.

Examples¶

Basic load from disk¶

import torch
from nicole import deserialize

payload = torch.load("tensor.tnsr", weights_only=True)
t = deserialize(payload)

Load to GPU¶

payload = torch.load("tensor.tnsr", weights_only=True)
t = deserialize(payload, device="cuda:0")

Round-trip in memory¶

from nicole import serialize, deserialize

payload = serialize(t)
t2 = deserialize(payload)