fhelium.serialization.safetensors

Single-file safetensors persistence for FHElium values.

FILE_FORMAT

constant View source

FILE_FORMAT = 'fhelium-value'

FILE_SCHEMA_VERSION

constant View source

FILE_SCHEMA_VERSION = 2

T

constant View source

T = TypeVar('T', bound=TensorResident)

ValueFileMetadata

class View source

ValueFileMetadata(file_schema_version: int, value_schema_version: int, value_type: str, nbytes: int, tensor_metadata: dict[str, dict[str, Any]], value_metadata: dict[str, Any])

Validated metadata inspectable without materializing tensor payloads.

Attributes

Name	Type	Default/value
file_schema_version	int
value_schema_version	int
value_type	str
nbytes	int
tensor_metadata	dict[str, dict[str, Any]]
value_metadata	dict[str, Any]

save_value

function View source

def save_value(value: TensorResident, path: str | os.PathLike[str], *, allow_secret: bool=False, overwrite: bool=False) -> ValueFileMetadata: ...

Atomically save one value to the caller-selected file path.

The file uses the versioned ValueEnvelope representation. Saving a secret key requires allow_secret=True. The file stores plaintext key data; callers must protect the destination and apply any required encryption.

inspect_value

function View source

def inspect_value(path: str | os.PathLike[str]) -> ValueFileMetadata: ...

Inspect one value file without materializing its tensor payloads.

load_value

function View source

def load_value(path: str | os.PathLike[str], *, device: torch.device | str='cpu', expected_type: type[T] | None=None) -> T: ...

Load one value from a caller-selected value-file path.

This is a file-codec operation: the caller owns path naming, replacement, and lifecycle. ArtifactStore.get is the separate repository operation for logical names, generations, checksums, and catalog transactions. Materialization defaults to CPU unless device selects another target.