Module model_loader 

Model Loader with ONNX Runtime Integration

SPARC Phase 3: Implementation

This module provides lazy loading, caching, and thread-safe access to ONNX models.

Features

• Lazy Loading: Models are only loaded when first requested
• Caching: Loaded models are cached for reuse
• Thread-Safe: Uses Arc + RwLock for concurrent access
• Registry Integration: Uses ModelRegistry for model discovery
• Graceful Error Handling: Comprehensive error messages

Usage Example

use llm_shield_models::{ModelLoader, ModelRegistry, ModelType, ModelVariant};
use std::sync::Arc;

// Create registry
let registry = ModelRegistry::from_file("models/registry.json")?;

// Create loader
let loader = ModelLoader::new(Arc::new(registry));

// Load model (lazy - only loads once)
let session = loader.load(ModelType::PromptInjection, ModelVariant::FP16).await?;

// Use session for inference...

Structs

LoaderStats

Statistics about loaded models

ModelConfig

Configuration for loading a model

ModelLoader

Model loader with lazy loading and caching

Enums

ModelType

Model type identifier

Type Aliases

Result

Result type alias