Struct QuantumEmbedding 

pub struct QuantumEmbedding {
    pub vocab_size: usize,
    pub dim: usize,
    pub angles: Vec<f32>,
}

Quantum embedding: maps vocab indices to density matrices.

Fields

vocab_size: usize

dim: usize

angles: Vec<f32>

Per-token rotation angles [vocab_size × dim]. Learnable. Each token gets dim angles that parameterize a unitary rotation.

Implementations

impl QuantumEmbedding

pub fn new(vocab_size: usize, dim: usize, seed: u64) -> Self

pub fn embed(&self, token: usize) -> DensityMatrixN

Embed a single token as a density matrix. Start with |k mod dim⟩, then apply parameterized rotations to spread the state across modes. The rotations are the learnable embedding.

pub fn embed_amplitude(&self, token: usize) -> Vec<Complex>

Embed a single token as an amplitude vector (Fock space representation).

Returns the dim-dimensional complex state vector |ψ⟩ WITHOUT forming the density matrix ρ = |ψ⟩⟨ψ|. This is the Fock space representation: dim amplitudes instead of dim² matrix entries.

BA-62: At dim=86, this returns 688 bytes instead of 59,168 bytes (86× smaller). The populations |ψ_k|² are identical to the diagonal of ρ = |ψ⟩⟨ψ|.

pub fn num_params(&self) -> usize

Number of learnable parameters.

Trait Implementations

impl Clone for QuantumEmbedding

fn clone(&self) -> QuantumEmbedding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.