/**
 * Configuration for a single ONNX model to be loaded.
 */
export interface OnnxModelConfig {
    /**
     * Unique identifier for the model (e.g., 'yolo-pose', 'arcface').
     * Used to retrieve the session via OnnxRuntimeService.getSession(name).
     */
    name: string;
    /**
     * Path to the ONNX model file, relative to process.cwd().
     * Example: 'models/yolo11l-pose.onnx'
     */
    path: string;
    /**
     * URL to download the model from if not present locally.
     * Typically a Hugging Face URL like:
     * 'https://huggingface.co/<user>/<repo>/resolve/main/<filename>'
     */
    downloadUrl?: string;
    /**
     * Load priority. Lower numbers load first.
     * Use to load smaller models first for faster initial availability.
     */
    priority: number;
    /**
     * If true, failure to load this model won't prevent app startup.
     * The model will be unavailable but other services can still function.
     */
    optional: boolean;
}
/**
 * Configuration options for the OnnxRuntimeModule.
 *
 * Note: The 'models' array is now optional since ModelManagerService
 * handles model loading from models.config.yaml. This parameter is kept
 * for backward compatibility but is no longer used.
 */
export interface OnnxRuntimeModuleOptions {
    /**
     * List of models to load during application bootstrap.
     * DEPRECATED: ModelManagerService now handles this from models.config.yaml
     * Kept for backward compatibility.
     */
    models?: OnnxModelConfig[];
    /**
     * Number of threads for intra-operation parallelism.
     * Controls parallelism within a single ONNX operation (e.g., matrix multiplication).
     * Default: 2 (conservative to avoid contention with @huggingface/transformers)
     */
    intraOpNumThreads?: number;
    /**
     * Number of threads for inter-operation parallelism.
     * Controls parallelism between independent operations in the graph.
     * Default: 1 (sequential to avoid thread contention)
     */
    interOpNumThreads?: number;
    /**
     * Level of graph optimization to apply.
     * - 'disabled': No optimizations
     * - 'basic': Basic optimizations (constant folding)
     * - 'extended': Extended optimizations (more aggressive)
     * - 'all': All available optimizations
     * Default: 'all'
     */
    graphOptimizationLevel?: "disabled" | "basic" | "extended" | "all";
}
/**
 * Default options for OnnxRuntimeModule.
 */
export declare const DEFAULT_ONNX_RUNTIME_OPTIONS: Partial<OnnxRuntimeModuleOptions>;
//# sourceMappingURL=onnx-runtime.config.d.ts.map