preload.js

/**
 * preload.js — Downloads and caches the model during docker build.
 */
import { pipeline, env } from '@xenova/transformers';
import path from 'path';
import { fileURLToPath } from 'url';

const __dirname = path.dirname(fileURLToPath(import.meta.url));

// Disable multi-threading (not available in Alpine build context)
import ort from 'onnxruntime-web';
ort.env.wasm.numThreads = 1;

const MODEL = process.env.MODEL || 'Xenova/bge-small-en-v1.5';

env.cacheDir = path.join(__dirname, '.cache');
env.allowLocalModels = true;

console.log(`Preloading model: ${MODEL}...`);
const start = Date.now();

const pipe = await pipeline('feature-extraction', MODEL);
const output = await pipe('test', { pooling: 'mean', normalize: true });

console.log(`Done in ${Date.now() - start}ms — dims: ${output.dims}`);
process.exit(0);