---
title: "什么是 Fine-Tuning？"
canonical: https://wavect.io/zh/glossary/fine-tuning/
language: zh
description: "在你自己的示例上继续训练模型以改变它的行为，这对于风格和格式是正确的工具，而对于注入最新事实则是错误的工具。"
image: "https://wavect.io/img/general/bak/open_graph_preview.jpg"
---

技术

# Fine-Tuning

在你自己的示例上继续训练模型以改变它的行为，这对于风格和格式是正确的工具，而对于注入最新事实则是错误的工具。

最近审阅: 2026-07-29 审阅人 [Kevin Riedl](/zh/team/kevin-riedl/) [wiki ↗](https://www.wikidata.org/wiki/Q139796365)

Fine-Tuning 拿一个预训练模型，在一组精选的你自己的示例上进一步训练它，调整模型权重，使它倾向于你所演示的行为。你用它来固化一致的语气、特定的输出格式、领域词汇，或基础模型处理得笨拙的某项任务。它改变的是模型如何回应，而不是它能访问哪些事实。

真正要紧的决策是 Fine-Tuning 与 RAG 之争，而人们常常把它弄反。RAG 通过从你的数据中检索，在运行时注入最新的、会变化的事实，所以答案只和你最后一次更新文档一样新。Fine-Tuning 教授持久的行为，但把知识冻结在训练时刻，所以当你的事实会变化时它就是错误的工具，它也不能可靠地阻止模型编造东西，因为 [幻觉](/zh/glossary/hallucination/) 是靠接地来减少的，而不是靠微调。经验法则：为形式而调，为事实而取。许多真实系统两者都用，一个经过微调、以你自家风格回答的模型，再由检索为实时数据提供依据。

成本现实没有过去那么吓人，但依然真实。你需要一个干净、带标注的数据集（通常是数百到数千个示例）、训练运行本身，以及每当基础模型改进或你的需求变化时重新微调的持续成本。最后这项成本是团队会忘记的那一项。Fine-Tune 不是一次性项目，而是一项维护承诺。

如果需要一条当前且低成本的实施路线，请阅读 [用 Unsloth 和免费 Colab 微调 Gemma 4 的实战指南](/zh/blog/fine-tune-gemma-4-free-unsloth-colab/) 。它涵盖浏览器试点、资源限制、评测、许可证、导出以及进入生产前的差距。

Fine-Tuning 彻底取胜的场景，举个实际例子：一家公司需要在数百万次调用中，让每一次模型回应都返回严格符合某个内部 schema 的 JSON，并带有一种特定的、简洁的自家风格。Prompting 大多数时候能做到，但偶尔一次格式错误的回应就会弄坏下游系统，而把整份风格指南和 schema 塞进每个 Prompt，又会在每次调用上烧掉 token。一次微调把格式和语气固化进模型权重，于是这些指令不必再在上下文里重复，在那个调用量级上，回应既更可靠、单次又更便宜。那就是甜蜜点：持久的行为、高调用量，以及一种 Prompting 始终无法稳定拿下的格式。

Fine-Tuning 什么时候彻底取胜？当 Prompting 加检索无法可靠地产出你所需的格式或行为，而你又有足够多高质量的示例去教它时。拿不准时，先把 Prompting 和 RAG 用尽，因为它们改起来更便宜。我们在 [人工智能](/zh/services/artificial-intelligence/) 下做出这个自建还是微调的判断。

// FAQ

## 常见问题

### Fine-Tuning 与 RAG，我该选哪个？

为形式而调，为事实而取。Fine-Tuning 改变模型的行为方式（语气、格式、任务）。RAG 在运行时注入最新事实。如果你的数据经常变化，你需要的是 RAG，而不是把知识冻结的 Fine-Tune。

### Fine-Tuning 要花多少钱？

比训练运行本身所暗示的更多。要为构建一个干净的带标注数据集、训练本身，以及每当基础模型或需求变化时重新微调的经常性成本做预算。维护是团队低估的那一部分。

### 什么时候 Fine-Tuning 是错误的选择？

当你需要答案中包含最新或频繁变化的事实时。Fine-Tuning 在训练时刻把知识固化进去。对于实时数据，使用检索。如果 Prompting 和 RAG 已经能更便宜地达到目的，它也是错误的选择。

相关术语

- [LLM](/zh/glossary/llm/)
- [RAG](/zh/glossary/rag/)
- [Hallucination](/zh/glossary/hallucination/)

相关服务

- [人工智能](/zh/services/artificial-intelligence/)

相关案例研究

- [Twinsoft AI](/zh/case-studies/twinsoft-ai/)

相关指南

- [如何为 MVP 选择技术栈](/zh/software-development-guide/how-to-choose-a-tech-stack-for-mvp/)

聊聊

## Structured Data

```json
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@id": "https://wavect.io/#organization",
      "@type": [
        "Organization",
        "ProfessionalService",
        "LocalBusiness"
      ],
      "employee": [
        {
          "@id": "https://wavect.io/team/kevin-riedl/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Kevin Riedl",
          "url": "https://wavect.io/team/kevin-riedl/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        },
        {
          "@id": "https://wavect.io/team/christof-jori/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Christof Jori",
          "url": "https://wavect.io/team/christof-jori/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        }
      ],
      "founder": [
        {
          "@id": "https://wavect.io/team/kevin-riedl/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Kevin Riedl",
          "url": "https://wavect.io/team/kevin-riedl/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        },
        {
          "@id": "https://wavect.io/team/christof-jori/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Christof Jori",
          "url": "https://wavect.io/team/christof-jori/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        }
      ],
      "legalRepresentative": [
        {
          "@id": "https://wavect.io/team/kevin-riedl/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Kevin Riedl",
          "url": "https://wavect.io/team/kevin-riedl/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        },
        {
          "@id": "https://wavect.io/team/christof-jori/#person",
          "@type": "Person",
          "jobTitle": "Managing Director",
          "name": "Christof Jori",
          "url": "https://wavect.io/team/christof-jori/",
          "worksFor": {
            "@id": "https://wavect.io/#organization",
            "@type": [
              "Organization",
              "ProfessionalService",
              "LocalBusiness"
            ]
          }
        }
      ],
      "name": "Wavect GmbH",
      "subjectOf": {
        "@id": "https://wavect.io/verified-claims.json#dataset",
        "@type": "Dataset",
        "creator": {
          "@id": "https://wavect.io/#organization",
          "@type": [
            "Organization",
            "ProfessionalService",
            "LocalBusiness"
          ]
        },
        "description": "A machine-readable registry of quantitative and qualitative claims published by Wavect, with review dates, localized page appearances and public third-party citations where available.",
        "inLanguage": "en",
        "isAccessibleForFree": true,
        "license": "https://creativecommons.org/licenses/by/4.0/",
        "name": "Wavect verified publication claims",
        "url": "https://wavect.io/verified-claims.json"
      },
      "url": "https://wavect.io/"
    },
    {
      "@id": "https://wavect.io/team/kevin-riedl/#person",
      "@type": "Person",
      "jobTitle": "Managing Director",
      "name": "Kevin Riedl",
      "sameAs": [
        "https://www.wikidata.org/wiki/Q139796365",
        "https://www.linkedin.com/in/wsdt",
        "https://github.com/wsdt"
      ],
      "url": "https://wavect.io/team/kevin-riedl/",
      "worksFor": {
        "@id": "https://wavect.io/#organization",
        "@type": [
          "Organization",
          "ProfessionalService",
          "LocalBusiness"
        ]
      }
    },
    {
      "@id": "https://wavect.io/team/christof-jori/#person",
      "@type": "Person",
      "jobTitle": "Managing Director",
      "name": "Christof Jori",
      "sameAs": [
        "https://www.wikidata.org/wiki/Q139796367",
        "https://www.linkedin.com/in/jocr77/",
        "https://github.com/jo-chris"
      ],
      "url": "https://wavect.io/team/christof-jori/",
      "worksFor": {
        "@id": "https://wavect.io/#organization",
        "@type": [
          "Organization",
          "ProfessionalService",
          "LocalBusiness"
        ]
      }
    },
    {
      "@id": "https://wavect.io/#website",
      "@type": "WebSite",
      "inLanguage": [
        "en",
        "de",
        "es",
        "zh"
      ],
      "name": "Wavect",
      "potentialAction": {
        "@type": "SearchAction",
        "query-input": "required name=search_term_string",
        "target": {
          "@type": "EntryPoint",
          "urlTemplate": "https://wavect.io/search/?q={search_term_string}"
        }
      },
      "publisher": {
        "@id": "https://wavect.io/#organization",
        "@type": [
          "Organization",
          "ProfessionalService",
          "LocalBusiness"
        ]
      },
      "url": "https://wavect.io/"
    }
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@id": "https://wavect.io/zh/glossary/fine-tuning/#term",
      "@type": "DefinedTerm",
      "alternateName": [],
      "description": "在你自己的示例上继续训练模型以改变它的行为，这对于风格和格式是正确的工具，而对于注入最新事实则是错误的工具。",
      "inDefinedTermSet": "https://wavect.io/zh/glossary/#termset",
      "name": "Fine-Tuning",
      "termCode": "fine-tuning",
      "url": "https://wavect.io/zh/glossary/fine-tuning/"
    },
    {
      "@id": "https://wavect.io/zh/glossary/fine-tuning/#webpage",
      "@type": "WebPage",
      "about": {
        "@id": "https://wavect.io/#organization",
        "@type": [
          "Organization",
          "ProfessionalService",
          "LocalBusiness"
        ]
      },
      "author": {
        "@id": "https://wavect.io/team/kevin-riedl/#person",
        "@type": "Person",
        "name": "Kevin Riedl",
        "url": "https://wavect.io/team/kevin-riedl/"
      },
      "dateModified": "2026-07-29",
      "description": "在你自己的示例上继续训练模型以改变它的行为，这对于风格和格式是正确的工具，而对于注入最新事实则是错误的工具。",
      "headline": "Fine-Tuning",
      "inLanguage": "zh",
      "isPartOf": {
        "@id": "https://wavect.io/#website",
        "@type": "WebSite"
      },
      "lastReviewed": "2026-07-29",
      "mainEntity": {
        "@id": "https://wavect.io/zh/glossary/fine-tuning/#term"
      },
      "mentions": [
        {
          "@id": "https://wavect.io/#organization",
          "@type": [
            "Organization",
            "ProfessionalService",
            "LocalBusiness"
          ]
        },
        {
          "@type": "Service",
          "name": "人工智能",
          "url": "https://wavect.io/zh/services/artificial-intelligence/"
        }
      ],
      "name": "Fine-Tuning",
      "reviewedBy": {
        "@id": "https://wavect.io/team/christof-jori/#person",
        "@type": "Person",
        "name": "Christof Jori",
        "url": "https://wavect.io/team/christof-jori/"
      },
      "speakable": {
        "@type": "SpeakableSpecification",
        "cssSelector": [
          ".gls-single__h1",
          ".gls-single__tldr"
        ]
      },
      "url": "https://wavect.io/zh/glossary/fine-tuning/"
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [
        {
          "@type": "ListItem",
          "item": "https://wavect.io/zh/",
          "name": "首页",
          "position": 1
        },
        {
          "@type": "ListItem",
          "item": "https://wavect.io/zh/glossary/",
          "name": "技术、产品与合作模式术语表",
          "position": 2
        },
        {
          "@type": "ListItem",
          "item": "https://wavect.io/zh/glossary/fine-tuning/",
          "name": "Fine-Tuning",
          "position": 3
        }
      ]
    }
  ]
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "为形式而调，为事实而取。Fine-Tuning 改变模型的行为方式（语气、格式、任务）。RAG 在运行时注入最新事实。如果你的数据经常变化，你需要的是 RAG，而不是把知识冻结的 Fine-Tune。"
      },
      "name": "Fine-Tuning 与 RAG，我该选哪个？"
    },
    {
      "@type": "Question",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "比训练运行本身所暗示的更多。要为构建一个干净的带标注数据集、训练本身，以及每当基础模型或需求变化时重新微调的经常性成本做预算。维护是团队低估的那一部分。"
      },
      "name": "Fine-Tuning 要花多少钱？"
    },
    {
      "@type": "Question",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "当你需要答案中包含最新或频繁变化的事实时。Fine-Tuning 在训练时刻把知识固化进去。对于实时数据，使用检索。如果 Prompting 和 RAG 已经能更便宜地达到目的，它也是错误的选择。"
      },
      "name": "什么时候 Fine-Tuning 是错误的选择？"
    }
  ],
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      ".faq-question",
      ".faq-answer"
    ]
  }
}
```