dist/core/onnx-embedder.js

"use strict";
/**
 * ONNX WASM Embedder - Semantic embeddings for hooks
 *
 * Provides real transformer-based embeddings using all-MiniLM-L6-v2
 * running in pure WASM (no native dependencies).
 *
 * Uses bundled ONNX WASM files from src/core/onnx/
 *
 * Features:
 * - 384-dimensional semantic embeddings
 * - Real semantic understanding (not hash-based)
 * - Cached model loading (downloads from HuggingFace on first use)
 * - Batch embedding support
 * - Optional parallel workers for 3.8x batch speedup
 */
var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
    if (k2 === undefined) k2 = k;
    var desc = Object.getOwnPropertyDescriptor(m, k);
    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
      desc = { enumerable: true, get: function() { return m[k]; } };
    }
    Object.defineProperty(o, k2, desc);
}) : (function(o, m, k, k2) {
    if (k2 === undefined) k2 = k;
    o[k2] = m[k];
}));
var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
    Object.defineProperty(o, "default", { enumerable: true, value: v });
}) : function(o, v) {
    o["default"] = v;
});
var __importStar = (this && this.__importStar) || (function () {
    var ownKeys = function(o) {
        ownKeys = Object.getOwnPropertyNames || function (o) {
            var ar = [];
            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
            return ar;
        };
        return ownKeys(o);
    };
    return function (mod) {
        if (mod && mod.__esModule) return mod;
        var result = {};
        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
        __setModuleDefault(result, mod);
        return result;
    };
})();
Object.defineProperty(exports, "__esModule", { value: true });
exports.OnnxEmbedder = void 0;
exports.isOnnxAvailable = isOnnxAvailable;
exports.initOnnxEmbedder = initOnnxEmbedder;
exports.embed = embed;
exports.embedBatch = embedBatch;
exports.similarity = similarity;
exports.cosineSimilarity = cosineSimilarity;
exports.getDimension = getDimension;
exports.isReady = isReady;
exports.getStats = getStats;
exports.shutdown = shutdown;
const path = __importStar(require("path"));
const fs = __importStar(require("fs"));
const url_1 = require("url");
const module_1 = require("module");
// Set up ESM-compatible require for WASM module (fixes Windows/ESM compatibility)
// The WASM bindings use module.require for Node.js crypto, this provides a fallback
if (typeof globalThis !== 'undefined' && !globalThis.__ruvector_require) {
    try {
        // In ESM context, use createRequire with __filename
        globalThis.__ruvector_require = (0, module_1.createRequire)(__filename);
    }
    catch {
        // Fallback: require should be available in CommonJS
        try {
            globalThis.__ruvector_require = require;
        }
        catch {
            // Neither available - WASM will fall back to crypto.getRandomValues
        }
    }
}
// Force native dynamic import (avoids TypeScript transpiling to require)
// eslint-disable-next-line @typescript-eslint/no-implied-eval
const dynamicImport = new Function('specifier', 'return import(specifier)');
// Capability detection
let simdAvailable = false;
let parallelAvailable = false;
// Lazy-loaded module state
let wasmModule = null;
let embedder = null;
let parallelEmbedder = null;
let loadError = null;
let loadPromise = null;
let isInitialized = false;
let parallelEnabled = false;
let parallelThreshold = 4;
// Default model
const DEFAULT_MODEL = 'all-MiniLM-L6-v2';
/**
 * Check if ONNX embedder is available (bundled files exist)
 */
function isOnnxAvailable() {
    try {
        const pkgPath = path.join(__dirname, 'onnx', 'pkg', 'ruvector_onnx_embeddings_wasm.js');
        return fs.existsSync(pkgPath);
    }
    catch {
        return false;
    }
}
/**
 * Check if parallel workers are available (npm package installed)
 */
async function detectParallelAvailable() {
    try {
        await dynamicImport('ruvector-onnx-embeddings-wasm/parallel');
        parallelAvailable = true;
        return true;
    }
    catch {
        parallelAvailable = false;
        return false;
    }
}
/**
 * Check if SIMD is available (from WASM module)
 */
function detectSimd() {
    try {
        if (wasmModule && typeof wasmModule.simd_available === 'function') {
            simdAvailable = wasmModule.simd_available();
            return simdAvailable;
        }
    }
    catch { }
    return false;
}
/**
 * Try to load ParallelEmbedder from npm package (optional)
 */
async function tryInitParallel(config) {
    // Skip if explicitly disabled
    if (config.enableParallel === false)
        return false;
    // For 'auto' or true, try to initialize
    try {
        const parallelModule = await dynamicImport('ruvector-onnx-embeddings-wasm/parallel');
        const { ParallelEmbedder } = parallelModule;
        parallelEmbedder = new ParallelEmbedder({
            numWorkers: config.numWorkers,
        });
        await parallelEmbedder.init(config.modelId || DEFAULT_MODEL);
        parallelThreshold = config.parallelThreshold || 4;
        parallelEnabled = true;
        parallelAvailable = true;
        console.error(`Parallel embedder ready: ${parallelEmbedder.numWorkers} workers, SIMD: ${simdAvailable}`);
        return true;
    }
    catch (e) {
        parallelAvailable = false;
        if (config.enableParallel === true) {
            // Only warn if explicitly requested
            console.error(`Parallel embedder not available: ${e.message}`);
        }
        return false;
    }
}
/**
 * Initialize the ONNX embedder (downloads model if needed)
 */
async function initOnnxEmbedder(config = {}) {
    if (isInitialized)
        return true;
    if (loadError)
        throw loadError;
    if (loadPromise) {
        await loadPromise;
        return isInitialized;
    }
    loadPromise = (async () => {
        try {
            // Paths to bundled ONNX files
            const pkgPath = path.join(__dirname, 'onnx', 'pkg', 'ruvector_onnx_embeddings_wasm.js');
            const loaderPath = path.join(__dirname, 'onnx', 'loader.js');
            if (!fs.existsSync(pkgPath)) {
                throw new Error('ONNX WASM files not bundled. The onnx/ directory is missing.');
            }
            // Convert paths to file:// URLs for cross-platform ESM compatibility (Windows fix)
            const pkgUrl = (0, url_1.pathToFileURL)(pkgPath).href;
            const loaderUrl = (0, url_1.pathToFileURL)(loaderPath).href;
            // Dynamic import of bundled modules using file:// URLs
            wasmModule = await dynamicImport(pkgUrl);
            // Initialize WASM module (loads the .wasm file)
            const wasmPath = path.join(__dirname, 'onnx', 'pkg', 'ruvector_onnx_embeddings_wasm_bg.wasm');
            if (wasmModule.default && typeof wasmModule.default === 'function') {
                // For bundler-style initialization, pass the wasm buffer
                const wasmBytes = fs.readFileSync(wasmPath);
                await wasmModule.default(wasmBytes);
            }
            const loaderModule = await dynamicImport(loaderUrl);
            const { ModelLoader } = loaderModule;
            // Create model loader with caching
            const modelLoader = new ModelLoader({
                cache: true,
                cacheDir: config.cacheDir || path.join(process.env.HOME || '/tmp', '.ruvector', 'models'),
            });
            // Load model (downloads from HuggingFace on first use)
            const modelId = config.modelId || DEFAULT_MODEL;
            console.error(`Loading ONNX model: ${modelId}...`);
            const { modelBytes, tokenizerJson, config: modelConfig } = await modelLoader.loadModel(modelId);
            // Create embedder with config
            const embedderConfig = new wasmModule.WasmEmbedderConfig()
                .setMaxLength(config.maxLength || modelConfig.maxLength || 256)
                .setNormalize(config.normalize !== false)
                .setPooling(0); // Mean pooling
            embedder = wasmModule.WasmEmbedder.withConfig(modelBytes, tokenizerJson, embedderConfig);
            // Detect SIMD capability
            detectSimd();
            console.error(`ONNX embedder ready: ${embedder.dimension()}d, SIMD: ${simdAvailable}`);
            isInitialized = true;
            // Determine if we should use parallel workers
            // - true: always enable
            // - false: never enable
            // - 'auto'/undefined: enable for long-running processes (MCP, servers), skip for CLI
            let shouldTryParallel = false;
            if (config.enableParallel === true) {
                shouldTryParallel = true;
            }
            else if (config.enableParallel === false) {
                shouldTryParallel = false;
            }
            else {
                // Auto-detect: check if running as CLI hook or long-running process
                const isCLI = process.argv[1]?.includes('cli.js') ||
                    process.argv[1]?.includes('bin/ruvector') ||
                    process.env.RUVECTOR_CLI === '1';
                const isMCP = process.env.MCP_SERVER === '1' ||
                    process.argv.some(a => a.includes('mcp'));
                const forceParallel = process.env.RUVECTOR_PARALLEL === '1';
                // Enable parallel for MCP/servers or if explicitly requested, skip for CLI
                shouldTryParallel = forceParallel || (isMCP && !isCLI);
            }
            if (shouldTryParallel) {
                await tryInitParallel(config);
            }
        }
        catch (e) {
            loadError = new Error(`Failed to initialize ONNX embedder: ${e.message}`);
            throw loadError;
        }
    })();
    await loadPromise;
    return isInitialized;
}
/**
 * Generate embedding for text
 */
async function embed(text) {
    if (!isInitialized) {
        await initOnnxEmbedder();
    }
    if (!embedder) {
        throw new Error('ONNX embedder not initialized');
    }
    const start = performance.now();
    const embedding = embedder.embedOne(text);
    const timeMs = performance.now() - start;
    return {
        embedding: Array.from(embedding),
        dimension: embedding.length,
        timeMs,
    };
}
/**
 * Generate embeddings for multiple texts
 * Uses parallel workers automatically for batches >= parallelThreshold
 */
async function embedBatch(texts) {
    if (!isInitialized) {
        await initOnnxEmbedder();
    }
    if (!embedder) {
        throw new Error('ONNX embedder not initialized');
    }
    const start = performance.now();
    // Use parallel workers for large batches
    if (parallelEnabled && parallelEmbedder && texts.length >= parallelThreshold) {
        const batchResults = await parallelEmbedder.embedBatch(texts);
        const totalTime = performance.now() - start;
        const dimension = parallelEmbedder.dimension || 384;
        return batchResults.map((emb) => ({
            embedding: Array.from(emb),
            dimension,
            timeMs: totalTime / texts.length,
        }));
    }
    // Sequential fallback
    const batchEmbeddings = embedder.embedBatch(texts);
    const totalTime = performance.now() - start;
    const dimension = embedder.dimension();
    const results = [];
    for (let i = 0; i < texts.length; i++) {
        const embedding = batchEmbeddings.slice(i * dimension, (i + 1) * dimension);
        results.push({
            embedding: Array.from(embedding),
            dimension,
            timeMs: totalTime / texts.length,
        });
    }
    return results;
}
/**
 * Calculate cosine similarity between two texts
 */
async function similarity(text1, text2) {
    if (!isInitialized) {
        await initOnnxEmbedder();
    }
    if (!embedder) {
        throw new Error('ONNX embedder not initialized');
    }
    const start = performance.now();
    const sim = embedder.similarity(text1, text2);
    const timeMs = performance.now() - start;
    return { similarity: sim, timeMs };
}
/**
 * Calculate cosine similarity between two embeddings
 */
function cosineSimilarity(a, b) {
    if (a.length !== b.length) {
        throw new Error('Embeddings must have same dimension');
    }
    let dotProduct = 0;
    let normA = 0;
    let normB = 0;
    for (let i = 0; i < a.length; i++) {
        dotProduct += a[i] * b[i];
        normA += a[i] * a[i];
        normB += b[i] * b[i];
    }
    const magnitude = Math.sqrt(normA) * Math.sqrt(normB);
    return magnitude === 0 ? 0 : dotProduct / magnitude;
}
/**
 * Get embedding dimension
 */
function getDimension() {
    return embedder ? embedder.dimension() : 384;
}
/**
 * Check if embedder is ready
 */
function isReady() {
    return isInitialized;
}
/**
 * Get embedder stats including SIMD and parallel capabilities
 */
function getStats() {
    return {
        ready: isInitialized,
        dimension: embedder ? embedder.dimension() : 384,
        model: DEFAULT_MODEL,
        simd: simdAvailable,
        parallel: parallelEnabled,
        parallelWorkers: parallelEmbedder?.numWorkers || 0,
        parallelThreshold,
    };
}
/**
 * Shutdown parallel workers (call on exit)
 */
async function shutdown() {
    if (parallelEmbedder) {
        await parallelEmbedder.shutdown();
        parallelEmbedder = null;
        parallelEnabled = false;
    }
}
// Export class wrapper for compatibility
class OnnxEmbedder {
    constructor(config = {}) {
        this.config = config;
    }
    async init() {
        return initOnnxEmbedder(this.config);
    }
    async embed(text) {
        const result = await embed(text);
        return result.embedding;
    }
    async embedBatch(texts) {
        const results = await embedBatch(texts);
        return results.map(r => r.embedding);
    }
    async similarity(text1, text2) {
        const result = await similarity(text1, text2);
        return result.similarity;
    }
    get dimension() {
        return getDimension();
    }
    get ready() {
        return isReady();
    }
}
exports.OnnxEmbedder = OnnxEmbedder;
exports.default = OnnxEmbedder;