{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigobjkggm7uexc3f5dwcvjcetd6m7endbgqdfhsa2kah4lqejfbeq",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mpnvmborlli2"
  },
  "path": "/t/local-llm-on-macbook-m5-pro-totally-new-to-this/177286#post_6",
  "publishedAt": "2026-07-02T10:46:01.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "Yes. :\n\n* * *\n\nFor that specific “not in the docs” test, I would turn web search off first.\n\nThe point of that test is not “can the model find the answer somehow?” The point is:\n\n> If the answer is not in the attached Knowledge/RAG source, does the model admit that, or does it guess?\n\nSo I would test in separate passes:\n\n  1. **Knowledge-only test**\n\n     * Web search off.\n\n     * Only the small test document / Knowledge collection available.\n\n     * Expected behavior: if the answer is not in the docs, the model should say so.\n\n  2. **Web-search test**\n\n     * Web search on.\n\n     * Ask a question where current outside information is useful.\n\n     * Expected behavior: the model should use web results and ideally say where the information came from.\n\n  3. **Hybrid test**\n\n     * Knowledge + web search both available.\n\n     * This is closer to real use, but harder to debug because you now have two retrieval sources.\n\n\n\n\nFor the first RAG smoke test, keep it boring and isolated. If web search is on, a correct answer may come from the web rather than from your Knowledge collection, and then you have not really tested whether RAG is working.\n\nOpen WebUI’s RAG and Web Search features are powerful, but I would test them separately before combining them.",
  "title": "Local LLM on MacBook M5 Pro - Totally New to This!"
}