Struct InterpretabilityEngine 

pub struct InterpretabilityEngine {
    pub model_info: ModelInfo,
    /* private fields */
}

The main interpretability engine

Fields

model_info: ModelInfo

Model dimensions

Implementations

impl InterpretabilityEngine

pub fn new(model_info: ModelInfo) -> Self

Create a new interpretability engine

pub fn record_activations( &mut self, input: &str, layer_activations: HashMap<usize, Vec<f32>>, attention_patterns: HashMap<(usize, usize), Vec<f32>>, )

Record activations for analysis

pub fn analyze_attention_head( &self, layer: usize, head: usize, ) -> Option<AttentionHead>

Analyze attention head patterns

pub fn discover_circuits(&mut self) -> Vec<Circuit>

Discover computational circuits

pub fn attribute_features(&self, tokens: Vec<String>) -> FeatureAttribution

Perform feature attribution for a decision

pub fn patch_activation( &mut self, layer: usize, position: usize, new_value: f32, ) -> ActivationPatch

Apply activation patching

pub fn probe_for_concept(&mut self, layer: usize, concept: &str) -> ProbeResult

Run probing classifier

pub fn analyze_safety(&mut self, input: &str) -> SafetyAnalysis

Perform comprehensive safety analysis

pub fn get_stats(&self) -> &InterpretabilityStats

Get engine statistics

pub fn get_circuits(&self) -> &[Circuit]

Get all discovered circuits

pub fn label_neuron( &mut self, layer: usize, position: usize, features: Vec<String>, )

Label a neuron with detected features

pub fn find_concept_neurons(&self, concept: &str) -> Vec<(usize, usize)>

Find neurons that respond to a specific concept

pub fn export_report(&self) -> InterpretabilityReport

Export interpretability report

Trait Implementations

impl Debug for InterpretabilityEngine

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

Formats the value using the given formatter.