FHElium

Appearance

fhelium.backend.ckks.codec._embedding ​

CKKS slot and coefficient embeddings for fixed-context codecs.

The functions in this module implement the generator-3 and generator-5 slot orders, the conjugate-symmetric negacyclic FFT embedding, and stochastic coefficient quantization. They operate on tensors and store no runtime state.

make_slot_tensor ​

function View source

python
def make_slot_tensor(message: Sequence[object] | torch.Tensor | complex | float | int, num_slots: int, device: torch.device) -> torch.Tensor: ...

Return a dense [*batch, slot] tensor with a full slot axis.

inverse_embed_slots ​

function View source

python
def inverse_embed_slots(message: torch.Tensor, *, pre: torch.Tensor, twister: torch.Tensor) -> torch.Tensor: ...

Map ordered slots to real coefficients using supplied embedding tables.

embed_coefficients ​

function View source

python
def embed_coefficients(coefficients: torch.Tensor, *, post: torch.Tensor, skewer: torch.Tensor) -> torch.Tensor: ...

Map real coefficients to the selected complex CKKS slot order.

encode_slots ​

function View source

python
def encode_slots(message: torch.Tensor, *, pre: torch.Tensor, twister: torch.Tensor, rounding_state: torch.Tensor, scale: float) -> torch.Tensor: ...

Stochastically quantize the inverse slot embedding with a live stream.

decode_slots ​

function View source

python
def decode_slots(coefficients: torch.Tensor, *, post: torch.Tensor, skewer: torch.Tensor, scale: float) -> torch.Tensor: ...

Recover ordered complex slots from scaled coefficients.