rag-ultimate

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.getEmbeddingModel = getEmbeddingModel;
exports.clearEmbeddingModelCache = clearEmbeddingModelCache;
let cachedModel = null;
// ── Loop prevention: concurrent-load deduplication ────────────────────────
// Without this, N parallel preprocess() calls that all observe cachedModel===null
// each race to call embedding.model(), launching N simultaneous load operations.
// Only the last one's result would survive (stored to cachedModel), the others
// are silently discarded — but the wasted round-trips can stall the pipeline
// long enough for a timeout, which the caller treats as a failure, which triggers
// an immediate retry, creating a tight retry loop.
// Sharing one in-flight Promise collapses all concurrent callers onto a single
// load operation and eliminates that source of looping.
let loadingPromise = null;
async function getEmbeddingModel(ctl) {
    if (cachedModel !== null)
        return cachedModel;
    // Return the existing in-flight promise so concurrent callers share one load.
    if (loadingPromise !== null)
        return loadingPromise;
    loadingPromise = ctl.client.embedding
        .model("nomic-ai/nomic-embed-text-v1.5-GGUF", { signal: ctl.abortSignal })
        .then((model) => {
        cachedModel = model;
        loadingPromise = null;
        return model;
    })
        .catch((error) => {
        // Reset both slots so the next caller retries from a clean state rather
        // than perpetually awaiting a rejected promise.
        cachedModel = null;
        loadingPromise = null;
        throw error;
    });
    return loadingPromise;
}
/** Force re-resolution of the embedding model (e.g. after a hot-reload). */
function clearEmbeddingModelCache() {
    cachedModel = null;
    loadingPromise = null;
}