fhelium.experimental.bootstrap.structural

Centered modulus raising and its integer-RNS reference helpers.

reference_centered_basis_extend

function View source

def reference_centered_basis_extend(residues: torch.Tensor, *, source_moduli: Sequence[int], target_moduli: Sequence[int], centered: bool=True) -> torch.Tensor: ...

Reconstruct source integers and reduce them into a target RNS basis.

This deliberately slow implementation is a test oracle for the native mixed-radix path. It applies ordinary CRT with Python integers to every coefficient. When centered is true, representatives greater than half the source modulus product are interpreted as negative before reduction into each target modulus.

Parameters

• residues: Tensor shaped [..., source_row, coefficient].
• source_moduli: Modulus corresponding to each source row.
• target_moduli: Moduli of the desired output rows.
• centered: Choose signed half-interval representatives instead of least nonnegative representatives.

Returns

Tensor on the original device with shape [..., target_row, coefficient] and the same integer dtype.

ModRaisedCiphertext

class View source

ModRaisedCiphertext(ciphertext: Ciphertext, source_depth: int, source_prime_ids: tuple[int, ...], source_modulus_bits: int, scale: float)

A ciphertext plus the depleted Q basis extended by ModRaise.

Attributes

Name	Type	Default/value
ciphertext	Ciphertext
source_depth	int
source_prime_ids	tuple[int, ...]
source_modulus_bits	int
scale	float

from_engine

method

def from_engine(engine: Engine, ciphertext: Ciphertext, *, source_depth: int, scale: float) -> ModRaisedCiphertext: ...

Attach the source basis needed by the first raised transform.