Function ensureModelLoaded

src/core/reranker.ts:53–112

()

}

async function ensureModelLoaded(): Promise<void> {
  if (cachedModel && cachedTokenizer) return;
  if (initPromise) return initPromise;

  initPromise = (async () => {
    const { AutoTokenizer, AutoModelForSequenceClassification, env } =
      await import('@huggingface/transformers');

    console.error(`[reranker] Loading cross-encoder: ${DEFAULT_RERANKER_MODEL}`);
    console.error('[reranker] (First run will download the model - this may take a moment)');

    try {
      cachedTokenizer = await AutoTokenizer.from_pretrained(DEFAULT_RERANKER_MODEL);
      cachedModel = await AutoModelForSequenceClassification.from_pretrained(
        DEFAULT_RERANKER_MODEL,
        {
          dtype: 'q8',
          // Limit ONNX Runtime to half cores by default — prevents system freeze during indexing.
          session_options: {
            intraOpNumThreads: Math.max(1, Math.floor(os.cpus().length / 2)),
            interOpNumThreads: 1
          }
        }
      );
      rerankerHealth = 'fallback'; // loaded but not yet triggered
      console.error('[reranker] Cross-encoder loaded successfully');
    } catch (err) {
      const msg = err instanceof Error ? err.message : String(err);
      const isCorrupt = /protobuf|parse|corrupt/i.test(msg);
      resetLoadedState();

      if (isCorrupt) {
        // Corrupted cache is recoverable in-session once the bad model files are removed.
        console.error(`[reranker] Cache corruption detected: ${msg}`);
        console.error('[reranker] Clearing corrupted cache. Next call will re-download.');
        try {
          const cacheDir = env.cacheDir ?? null;
          if (cacheDir) {
            const { rmSync, existsSync } = await import('fs');
            const modelCacheDir = `${cacheDir}/Xenova/ms-marco-MiniLM-L-6-v2`;
            if (existsSync(modelCacheDir)) {
              rmSync(modelCacheDir, { recursive: true, force: true });
              console.error('[reranker] Corrupted cache cleared. Will re-download on next call.');
            }
          }
        } catch {
          // Cache clear is best-effort
        }
        rerankerHealth = 'unavailable';
        throw err;
      }

      // Transient error (network, timeout, etc.) — allow retry on next call.
      rerankerHealth = 'unavailable';
      throw err;
    }
  })();