{"version":"1.0","generated_at":"2026-11-01T14:47:12.821747","id":3683,"slug":"figma","title":"Figma 讲解如何训练专用画布编辑模型","summary":"Figma 公开其自研 Figma 编辑模型的训练路径：先用超过 25 万条合成样本做 SFT，把早期 Llama 3.3 70B 在编辑任务上推到接近当时顶级模型，随后转向 RL，并结合确定性校验器与视觉评委做评测。","abstract":"Figma 讲解如何训练专用画布编辑模型 Figma 公开其自研 Figma 编辑模型的训练路径：先用超过 25 万条合成样本做 SFT，把早期 Llama 3.3 70B 在编辑任务上推到接近当时顶级模型，随后转向 RL，并结合确定性校验器与视觉评委做评测。 Figma 公开其自研 Figma 编辑模型的训练路径：先用超过 25 万条合成样本做 SFT，把早期 Llama 3.3 70B 在编辑任务上推到接近当时顶级模型，随后转向 RL，并结合确定性校验器与视觉评委做评测。 原贴提到：Figma 介绍其为 agent 自研 Figma 编辑模型的训练方法，先用超过 25 万条合成样本做 SFT，让早期 Llama 3.3 70 来源：figma.com Figma 介绍其为 agent 自研 Figma 编辑模型的训练方法，先用超过 25 万条合成样本做 SFT，让早期 Llama 3.3 70B 在编辑任务上接近当时顶级模型，后转向 RL 并结合确定性校验器与视觉评委做评测。 这是什么信号 Figma 把自己训练编辑模型的路径公开讲了出来：第一步是监督微调，用超过 25 万条合成样本教模型“怎么改画布”，效果是让早期的 Llama 3.3 70B 在编辑任务上做到接近当时顶级模型的水平；第二步转向强化学习，同时在评测环节引入确定性校验器和视觉评委两套机制。这条信号的真正含义不是“Figma 也做模型了”，而是设计工具这类强结构化、强视觉反馈的垂直场景，正在被拆解成一个可训练的专用任务集。 为什么重要 通用模型的通用能力在画布编辑上并不天然好用，因为编辑动作有明确的可执行约…","access_level":"member","access_label":"免费阅读","access_mode":"full","is_preview":false,"canonical_url":"https://opc.beizhux.com/content/3683/figma","html_url":"https://opc.beizhux.com/content/3683/figma","json_url":"https://opc.beizhux.com/content/3683/figma.json","published_at":"2026-10-30T05:00:00","updated_at":"2026-11-01T08:45:22","category":{"slug":"ai-skills","name":"AI技能杠杆"},"source":{"site":"figma.com","author":"Figma AI / Agent / Make","url":"https://www.figma.com/blog/training-a-figma-fluent-editing-model/"},"tags":["Agent","AI Agent","Figma","SFT","合成数据","强化学习","模型训练","画布编辑","视觉评委","设计工具","评测"],"topics":[{"slug":"ai-tools","name":"AI工具","url":"https://opc.beizhux.com/topics/ai-tools","reason":"这篇内容命中「工具、模型」等主题信号。"},{"slug":"agent-workflow","name":"Agent工作流","url":"https://opc.beizhux.com/topics/agent-workflow","reason":"这篇内容命中「Agent、工作流」等主题信号。"},{"slug":"ai-super-individual","name":"AI超级个体","url":"https://opc.beizhux.com/topics/ai-super-individual","reason":"这篇内容命中「技能、学习」等主题信号。"}],"keywords":["AI技能杠杆","AI工具","Agent工作流","AI超级个体","工具","自动化","模型","Cursor","Claude","Agent","智能体","工作流"],"questions":[{"question":"Figma 讲解如何训练专用画布编辑模型主要讲什么？","answer":"Figma 公开其自研 Figma 编辑模型的训练路径：先用超过 25 万条合成样本做 SFT，把早期 Llama 3.3 70B 在编辑任务上推到接近当时顶级模型，随后转向 RL，并结合确定性校验器与视觉评委做评测。"},{"question":"这篇文章最值得关注的要点是什么？","answer":"Figma 公开其自研 Figma 编辑模型的训练路径：先用超过 25 万条合成样本做 SFT，把早期 Llama 3.3 70B 在编辑任务上推到接近当时顶级模型，随后转向 RL，并结合确定性校验器与视觉评委做评测。；原贴提到：Figma 介绍其为 agent 自研 Figma 编辑模型的训练方法，先用超过 25 万条合成样本做 SFT，让早期 Llama 3.3 70；来源：figma.com"},{"question":"这篇文章和哪些AI专题相关？","answer":"它适合放在AI工具、Agent工作流、AI超级个体专题里阅读。 关联原因：这篇内容命中「工具、模型」等主题信号。；这篇内容命中「Agent、工作流」等主题信号。；这篇内容命中「技能、学习」等主题信号。"},{"question":"阅读这篇文章建议先理解哪些关键词？","answer":"建议先理解AI工具、工具、自动化、模型、Cursor这些关键词，再结合正文判断工具、机会或风险是否值得进入自己的工作流。"}],"terms":[{"slug":"ai-tools-term","name":"AI工具","definition":"在AI觉醒星球里，「AI工具」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/ai-tools-term","path":"/glossary/ai-tools-term","json_url":"https://opc.beizhux.com/glossary/ai-tools-term.json"},{"slug":"tools","name":"工具","definition":"在AI觉醒星球里，「工具」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/tools","path":"/glossary/tools","json_url":"https://opc.beizhux.com/glossary/tools.json"},{"slug":"automation","name":"自动化","definition":"在AI觉醒星球里，「自动化」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/automation","path":"/glossary/automation","json_url":"https://opc.beizhux.com/glossary/automation.json"},{"slug":"ai-model","name":"模型","definition":"在AI觉醒星球里，「模型」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/ai-model","path":"/glossary/ai-model","json_url":"https://opc.beizhux.com/glossary/ai-model.json"},{"slug":"cursor","name":"Cursor","definition":"在AI觉醒星球里，「Cursor」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/cursor","path":"/glossary/cursor","json_url":"https://opc.beizhux.com/glossary/cursor.json"},{"slug":"claude","name":"Claude","definition":"在AI觉醒星球里，「Claude」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/claude","path":"/glossary/claude","json_url":"https://opc.beizhux.com/glossary/claude.json"},{"slug":"agent","name":"Agent","definition":"在AI觉醒星球里，「Agent」属于「Agent工作流」方向。聚合Agent、AI编程、Claude Code、Cursor、工作流自动化等内容，帮助你把智能体能力落到真实任务链路。 把智能体当成任务系统，而不是聊天窗口。","topic_slug":"agent-workflow","topic_name":"Agent工作流","topic_title":"Agent工作流：智能体、AI编程与自动化协作","topic_url":"https://opc.beizhux.com/topics/agent-workflow","topic_path":"/topics/agent-workflow","url":"https://opc.beizhux.com/glossary/agent","path":"/glossary/agent","json_url":"https://opc.beizhux.com/glossary/agent.json"},{"slug":"ai-agent","name":"智能体","definition":"在AI觉醒星球里，「智能体」属于「Agent工作流」方向。聚合Agent、AI编程、Claude Code、Cursor、工作流自动化等内容，帮助你把智能体能力落到真实任务链路。 把智能体当成任务系统，而不是聊天窗口。","topic_slug":"agent-workflow","topic_name":"Agent工作流","topic_title":"Agent工作流：智能体、AI编程与自动化协作","topic_url":"https://opc.beizhux.com/topics/agent-workflow","topic_path":"/topics/agent-workflow","url":"https://opc.beizhux.com/glossary/ai-agent","path":"/glossary/ai-agent","json_url":"https://opc.beizhux.com/glossary/ai-agent.json"}],"preview_hidden_sections":[],"related_items":[{"id":3726,"title":"datasette-llm 0.1a9：Datasette Agent 自动沿用 LLM 默认模型","url":"https://opc.beizhux.com/content/3726/datasette-llm-0-1a9","summary":"Simon Willison 发布 datasette-llm 0.1a9，这个版本会读取 LLM 已配置的默认模型，因此用户无需再为 Datasette Agent 显式指定默认模型即可直接上手；作者同时透露 Datasette Agent 的大幅升级正在准备中。","access_level":"member"},{"id":3624,"title":"Thorsten Ball 谈 Agent 工作法：横向打通构建、发版与生产观测","url":"https://opc.beizhux.com/content/3624/thorsten-ball-agent","summary":"邵猛转发并归纳 Thorsten Ball（AmpCode）的《How I Work with Agents》一文，核心观点是：只把 Agent 塞进编辑器这一格不会产生收益，要让 Agent 横向覆盖构建、发版、CI、生产日志、feature flag 和监控。","access_level":"member"},{"id":3616,"title":"Ethan Mollick 谈用 AI 要更敢想","url":"https://opc.beizhux.com/content/3616/ethan-mollick-ai","summary":"Ethan Mollick 提出用 AI 的通用方法：不断追问“怎样把这件事做得更有野心”，直到遇到能力边界，从而看清 AI 的前沿在哪里。","access_level":"member"}],"usage_policy":{"summarizable":true,"preferred_url":"https://opc.beizhux.com/content/3683/figma","private_fields_excluded":true}}