fhelium.native.runtime

Native Torch operator loading and runtime ABI validation.

NativeExtensionError

class View source

NativeExtensionError()

Bases: ImportError

Raised when native operators are unavailable or ABI-incompatible.

NativeStatus

class View source

NativeStatus(available: bool, reason: str, details: tuple[str, ...], binary_path: Path, manifest_path: Path, backends: tuple[str, ...])

Native-extension availability and diagnostic paths for this runtime.

available is true only after binary discovery, ABI-manifest validation, PyTorch loading, and FakeTensor registration succeed. reason and details explain the recorded result without triggering a new load.

Attributes

Name	Type	Default/value
available	bool
reason	str
details	tuple[str, ...]
binary_path	Path
manifest_path	Path
backends	tuple[str, ...]

native_available

function View source

def native_available() -> bool: ...

Return whether native operators loaded successfully for this runtime.

native_backend_available

function View source

def native_backend_available(backend: str) -> bool: ...

Return whether the loaded extension implements backend.

native_status

function View source

def native_status() -> NativeStatus: ...

Return the immutable native-extension status recorded during import.

require_native

function View source

def require_native() -> None: ...

Raise NativeExtensionError unless native operators are available.

require_native_backend

function View source

def require_native_backend(backend: str) -> None: ...

Raise unless the loaded extension implements backend.