{
  "source_file": "0930",
  "source_platform": "个人博客",
  "author": "Hamel Husain / Parlance Labs",
  "title_listed": "Is Fine-Tuning Still Valuable?",
  "keywords": "微调,价值,LLM",
  "domain": "大模型算法",
  "content_type": "观点文章",
  "is_paid_flag": "否",
  "url": "https://hamel.dev/blog/posts/fine_tuning_valuable.html",
  "publish_date": "2024-03",
  "corpus_stage": "预训练/SFT",
  "notes": "微调在LLM时代是否仍有价值",
  "id": "c405e84b11d6",
  "collected_at": "2026-10-02T14:28:49",
  "title": "Is Fine-Tuning Still Valuable? – Hamel’s Blog",
  "body_markdown": "Is Fine-Tuning Still Valuable?\nA reaction to a recent trend of disillusionment with fine-tuning.\nAuthor\nHamel Husain\nPublished\nMarch 27, 2024\nHere is my personal opinion about the questions I posed in\nthis tweet\n:\nThere are a growing number of voices expressing disillusionment with fine-tuning.\nI'm curious about the sentiment more generally. (I am withholding sharing my opinion rn).\nTweets below are from\n@mlpowered\n@abacaj\n@emollick\npic.twitter.com/cU0hCdubBU\n— Hamel Husain (\n@HamelHusain\n)\nMarch 26, 2024\nI think that fine-tuning is still very valuable in many situations. I’ve done some more digging and I find that people who say that fine-tuning isn’t useful are indeed often working on products where fine-tuning isn’t likely to be useful:\nThey are making developer tools - foundation models have been trained extensively on coding tasks.\nThey are building foundation models and testing for the most general cases. But the foundation models themselves are also being trained for the most general cases.\nThey are building a personal assistant that isn’t scoped to any type of domain or use case and is essentially similar to the same folks building foundation models.\nAnother common pattern is that people often say this in earlier stages of their product development. One sign that folks are in really early stages is that they don’t have a domain-specific eval harness.\nIt’s impossible to fine-tune effectively without an eval system which can lead to writing off fine-tuning if you haven’t completed this prerequisite.\nIt’s also impossible to improve your product without a good eval system in the long term, fine-tuning or not.\nYou should do as much prompt engineering as possible before you fine-tune. But not for reasons you would think! The reason for doing lots of prompt engineering is that it’s a great way to stress test your eval system!\nIf you find that prompt-engineering works fine (and you are systematically evaluating your product) then it’s fine to stop there. I’m a big believer in using the simplest approach to solving a problem. I just don’t think you should write off fine-tuning yet.\nExamples where I’ve seen fine-tuning work well\nGenerally speaking, fine-tuning works best to learn syntax, style and rules whereas techniques like RAG work best to supply the model with context or up-to-date facts.\nThese are some examples from companies I’ve worked with. Hopefully, we will be able to share more details soon.\nHoneycomb’s Natural Language Query Assistant\n- previously, the “programming manual” for the Honeycomb query language was being dumped into the prompt along with many examples. While this was OK, fine-tuning worked much better to allow the model to learn the syntax and rules of this niche domain-specific language.\nReChat’s Lucy\n- this is an AI real estate assistant integrated into an existing Real Estate CRM system. ReChat needs LLM responses to be provided in a very idiosyncratic format that weaves together structured and unstructured data to allow the front end to render widgets, cards and other interactive elements dynamically into the chat interface. Fine-tuning was the key to making this work correctly.\nThis talk\nhas more details.\nP.S. Fine-tuning is not only limited to open or “small” models. There are lots of folks who have been fine-tuning GPT-3.5, such as\nPerplexity.AI:\nand\nCoCounsel\n, to name a few.",
  "metadata": {
    "title": "Is Fine-Tuning Still Valuable? – Hamel’s Blog",
    "description": "A reaction to a recent trend of disillusionment with fine-tuning.",
    "keywords": null,
    "author": "Hamel Husain",
    "og:type": "article",
    "og:title": "Is Fine-Tuning Still Valuable? – Hamel’s Blog",
    "og:description": "A reaction to a recent trend of disillusionment with fine-tuning.",
    "og:image": "https://hamel.dev/quarto.png",
    "og:site_name": "Hamel’s Blog",
    "og:locale": "en_US",
    "twitter:title": "Is Fine-Tuning Still Valuable? – Hamel’s Blog",
    "twitter:description": "A reaction to a recent trend of disillusionment with fine-tuning.",
    "twitter:image": "https://hamel.dev/quarto.png",
    "twitter:creator": "@HamelHusain",
    "twitter:site": "@HamelHusain",
    "twitter:card": "summary_large_image"
  },
  "paywall": {
    "listed_paid": false,
    "detected": false,
    "marker": "",
    "public_only": false,
    "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙"
  },
  "crawl": {
    "success": true,
    "status_code": 200,
    "error": ""
  },
  "code_blocks": [],
  "images": [],
  "audio": [],
  "videos": [
    {
      "url": "https://www.youtube.com/watch?v=B_DMMlDuJB0",
      "src": "https://www.youtube.com/watch?v=B_DMMlDuJB0",
      "status": "source_unavailable",
      "kind": "single_video",
      "error": "ERROR: [youtube] B_DMMlDuJB0: Video unavailable\n",
      "title": "",
      "retried_at": "2026-10-20T17:59:54Z"
    }
  ],
  "links": [
    {
      "href": "https://www.googletagmanager.com/ns.html?id=GTM-PKGWQMKL",
      "label": "",
      "type": "iframe"
    },
    {
      "href": "https://x.com/HamelHusain/status/1772426234032541962?s=20",
      "label": "this tweet",
      "type": "a"
    },
    {
      "href": "https://twitter.com/mlpowered?ref_src=twsrc%5Etfw",
      "label": "@mlpowered",
      "type": "a"
    },
    {
      "href": "https://twitter.com/abacaj?ref_src=twsrc%5Etfw",
      "label": "@abacaj",
      "type": "a"
    },
    {
      "href": "https://twitter.com/emollick?ref_src=twsrc%5Etfw",
      "label": "@emollick",
      "type": "a"
    },
    {
      "href": "https://t.co/cU0hCdubBU",
      "label": "pic.twitter.com/cU0hCdubBU",
      "type": "a"
    },
    {
      "href": "https://twitter.com/HamelHusain/status/1772426234032541962?ref_src=twsrc%5Etfw",
      "label": "March 26, 2024",
      "type": "a"
    },
    {
      "href": "https://www.honeycomb.io/blog/introducing-query-assistant",
      "label": "Honeycomb’s Natural Language Query Assistant",
      "type": "a"
    },
    {
      "href": "https://www.youtube.com/watch?v=B_DMMlDuJB0",
      "label": "ReChat’s Lucy",
      "type": "a"
    },
    {
      "href": "https://x.com/perplexity_ai/status/1695102998463009254?s=20",
      "label": "Perplexity.AI:",
      "type": "a"
    },
    {
      "href": "https://legal.thomsonreuters.com/blog/from-capability-to-confidence-how-cocounsel-is-redefining-legal-ai/",
      "label": "CoCounsel",
      "type": "a"
    }
  ],
  "repair_version": 2,
  "repaired_at": "2026-10-14T10:52:41Z",
  "source_url": "https://hamel.dev/blog/posts/fine_tuning_valuable.html",
  "papers": [],
  "actual_video_files": [],
  "body_chars": 3366,
  "expected_body_images": 0,
  "discovery_status": "ok",
  "delivery_status": "partial",
  "missing_reason": "source_unavailable",
  "raw_html": "attachments/c405e84b11d6/pages/source.html",
  "local_html": "attachments/c405e84b11d6/pages/article.html",
  "body_base_path": "attachments/c405e84b11d6/pages/",
  "body_markdown_path": "attachments/c405e84b11d6/pages/article.md"
}