Struct QuantumMultiHeadAttention 

pub struct QuantumMultiHeadAttention { /* private fields */ }

Quantum multi-head attention module

Implementations

impl QuantumMultiHeadAttention

pub fn new( num_heads: usize, model_dim: usize, attention_type: QuantumAttentionType, num_qubits: usize, ) -> Result<Self>

Create new quantum multi-head attention module

Examples found in repository

examples/quantum_transformer.rs (line 132)

fn attention_mechanisms_demo() -> Result<()> {
    println!("   Testing various quantum attention mechanisms...");

    let attention_types = vec![
        ("Full Quantum", QuantumAttentionType::FullQuantum),
        (
            "Hybrid Quantum-Classical",
            QuantumAttentionType::HybridQuantumClassical,
        ),
        (
            "Variational Quantum",
            QuantumAttentionType::VariationalQuantum,
        ),
        (
            "Quantum Enhanced Multi-Head",
            QuantumAttentionType::QuantumEnhancedMultiHead,
        ),
        (
            "Quantum Self-Attention",
            QuantumAttentionType::QuantumSelfAttention,
        ),
    ];

    for (name, attention_type) in attention_types {
        println!("\n   --- {} Attention ---", name);

        let attention = QuantumMultiHeadAttention::new(4, 256, attention_type, 8)?;
        println!(
            "   Created attention module: {} heads, {} model dim",
            4, 256
        ); // Fixed values since fields are private

        // Test forward pass
        let batch_size = 2;
        let seq_len = 10;
        let model_dim = 256;

        let query = Array3::from_shape_fn((batch_size, seq_len, model_dim), |(b, s, d)| {
            0.1 * (b as f64 + s as f64 * 0.1 + d as f64 * 0.01)
        });
        let key = query.clone();
        let value = query.clone();

        let attention_output = attention.forward(&query, &key, &value, None)?;

        println!(
            "   Attention output shape: {:?}",
            attention_output.output.dim()
        );
        println!(
            "   Attention weights shape: {:?}",
            attention_output.attention_weights.dim()
        );

        // Analyze quantum attention properties
        let quantum_info = &attention_output.quantum_info;
        let avg_entanglement = quantum_info.entanglement_matrix.mean().unwrap_or(0.0);
        let max_coherence = quantum_info
            .coherence_scores
            .iter()
            .cloned()
            .fold(f64::NEG_INFINITY, f64::max);

        println!("   Average entanglement: {:.4}", avg_entanglement);
        println!("   Maximum coherence: {:.4}", max_coherence);

        // Attention pattern analysis
        let attention_weights = &attention_output.attention_weights;
        let max_attention = attention_weights
            .iter()
            .cloned()
            .fold(f64::NEG_INFINITY, f64::max);
        let avg_attention = attention_weights.mean().unwrap_or(0.0);

        println!("   Max attention weight: {:.4}", max_attention);
        println!("   Average attention: {:.4}", avg_attention);

        // Check attention sparsity
        let sparsity = attention_weights.iter().filter(|&&x| x < 0.01).count() as f64
            / attention_weights.len() as f64;
        println!("   Attention sparsity: {:.1}%", sparsity * 100.0);
    }

    Ok(())
}

pub fn forward( &self, query: &Array3<f64>, key: &Array3<f64>, value: &Array3<f64>, attention_mask: Option<&Array3<bool>>, ) -> Result<AttentionOutput>

Forward pass through quantum multi-head attention

Examples found in repository

examples/quantum_transformer.rs (line 149)

fn attention_mechanisms_demo() -> Result<()> {
    println!("   Testing various quantum attention mechanisms...");

    let attention_types = vec![
        ("Full Quantum", QuantumAttentionType::FullQuantum),
        (
            "Hybrid Quantum-Classical",
            QuantumAttentionType::HybridQuantumClassical,
        ),
        (
            "Variational Quantum",
            QuantumAttentionType::VariationalQuantum,
        ),
        (
            "Quantum Enhanced Multi-Head",
            QuantumAttentionType::QuantumEnhancedMultiHead,
        ),
        (
            "Quantum Self-Attention",
            QuantumAttentionType::QuantumSelfAttention,
        ),
    ];

    for (name, attention_type) in attention_types {
        println!("\n   --- {} Attention ---", name);

        let attention = QuantumMultiHeadAttention::new(4, 256, attention_type, 8)?;
        println!(
            "   Created attention module: {} heads, {} model dim",
            4, 256
        ); // Fixed values since fields are private

        // Test forward pass
        let batch_size = 2;
        let seq_len = 10;
        let model_dim = 256;

        let query = Array3::from_shape_fn((batch_size, seq_len, model_dim), |(b, s, d)| {
            0.1 * (b as f64 + s as f64 * 0.1 + d as f64 * 0.01)
        });
        let key = query.clone();
        let value = query.clone();

        let attention_output = attention.forward(&query, &key, &value, None)?;

        println!(
            "   Attention output shape: {:?}",
            attention_output.output.dim()
        );
        println!(
            "   Attention weights shape: {:?}",
            attention_output.attention_weights.dim()
        );

        // Analyze quantum attention properties
        let quantum_info = &attention_output.quantum_info;
        let avg_entanglement = quantum_info.entanglement_matrix.mean().unwrap_or(0.0);
        let max_coherence = quantum_info
            .coherence_scores
            .iter()
            .cloned()
            .fold(f64::NEG_INFINITY, f64::max);

        println!("   Average entanglement: {:.4}", avg_entanglement);
        println!("   Maximum coherence: {:.4}", max_coherence);

        // Attention pattern analysis
        let attention_weights = &attention_output.attention_weights;
        let max_attention = attention_weights
            .iter()
            .cloned()
            .fold(f64::NEG_INFINITY, f64::max);
        let avg_attention = attention_weights.mean().unwrap_or(0.0);

        println!("   Max attention weight: {:.4}", max_attention);
        println!("   Average attention: {:.4}", avg_attention);

        // Check attention sparsity
        let sparsity = attention_weights.iter().filter(|&&x| x < 0.01).count() as f64
            / attention_weights.len() as f64;
        println!("   Attention sparsity: {:.1}%", sparsity * 100.0);
    }

    Ok(())
}

Trait Implementations

impl Clone for QuantumMultiHeadAttention

fn clone(&self) -> QuantumMultiHeadAttention

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantumMultiHeadAttention

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.