[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-post-training-unified-coord-sft-rlhf-distill":3,"topics-all":36,"news-related-e648a701-6b9d-4a4b-9122-d2a4afc8349b":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"e648a701-6b9d-4a4b-9122-d2a4afc8349b","后训练方法终于有了统一坐标系：SFT、RLHF、Distillation 到底在做什么？","后训练方法终于有了统一坐标系：SFT、RLHF、Distillation 到底在做什么？\n\n大模型后训练方法列表越来越长：SFT、RLHF、PPO、DPO、distillation、process supervision……从业者各有偏好，但整个领域长期缺乏统一坐标系来理解这些方法究竟在解决什么问题。2026年5月，一篇来自多所高校联合团队的研究论文提出：将所有后训练方法统一理解为「对模型行为的三层干预」。\n\n## 两个维度构建的统一坐标系\n\n论文按两个维度分类所有后训练方法。第一维度按轨迹来源划分「离线」（学习外部提供的轨迹）和「在线」（学习模型自身生成的 rollouts）。第二维度按干预目标划分「有效支持扩展」——让有用行为更容易被触及，以及「策略重塑」——在已可触及的区域内改进行为。\n\n在这个坐标系下，SFT 既可以是支持扩展也可以是策略重塑，取决于用的是谁的数据；偏好优化（DPO\u002FPPO）通常是离线的策略重塑；在线 RL 在模型自身生成的 state 上改进行为；而 Distillation 被重新理解为「行为整合」而非单纯的压缩——这个视角的转换是这篇论文最有价值的洞察之一。\n\n## 为什么这个框架重要\n\n它不只是一个描述性框架，更能用来诊断实际瓶颈。如果你的模型在某个任务上表现差，第一步应该问：是「行为不可及」（需要支持扩展），还是「行为可及但质量不够」（需要策略重塑）？不同的问题对应不同的方法选型，选错方向浪费的不只是计算资源，更是模型能力的天花板。\n\n论文更深层的结论指向一个更大的趋势：2026年的后训练进步越来越依赖「协调的系统设计」而非任何单一主导目标。靠堆 SFT 数据或调 RL 超参就能提升模型的时代正在过去，未来的能力提升将来自对整个后训练 pipeline 的系统性规划。\n\n这意味着，后训练将从「炼金术调参」时代走向「系统工程」时代。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.07941v2","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"38c1bb96-546a-426c-b2ae-883bbf650a62","en","A unified map of post-training: SFT, RLHF, distillation explained","arXiv 2604.07941v2 proposes a unified framework for understanding post-training methods — SFT, RLHF, and distillation. The framework maps all post-training methods onto a common coordinate system, clarifying what each method is actually optimizing, and enabling systematic comparison and combination.","post-training-unified-coord-sft-rlhf-distill","2026-05-30T05:12:00Z","2026-05-30T13:09:19.727700Z","2026-08-19T02:08:40.142862Z",true,"agent",163,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"fbf3ec38-2aba-4199-8e55-c56071ea6e24","CAT 让 LRM 不再「想太多」:把模型自我置信度变成推理长度调速器","cat-confidence-adaptive-thinking","2026-07-05T16:05:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"b571067a-9fa8-42bf-9431-98f26ac78e03","伯克利把LLM推理搬进SSD:KV缓存压缩15倍","llm-inference-in-flash-cim-ssd","2026-09-19T21:10:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"b667e52f-ec7d-4ca4-8d9e-1db81e1a5616","DeepSeek论文:890字节KV缓存的三层架构账","deepseek-v41-flash-kv-cache-paper","2026-09-18T15:10:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"1942b07b-f794-42b1-b944-ca6b32d4ae16","四大 AI 模型同日集体掉线:OpenAI\u002FClaude 官方确认,Gemini\u002FGrok 表面沉默","four-ai-models-overlapping-outage-sept-2026","2026-09-06T08:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"1311adb6-dc19-41a7-a188-6760d9e53672","HF Summer 2026 报告:13 个下载量 Top 25 模型是 2022 年的老面孔","hugging-face-summer-2026-attention-adoption","2026-08-24T08:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"31f3215c-0892-419d-a610-fe815cc60bbe","GPT-5.6 降价 80% 把竞争拉进「同等智能成本」：DeepSeek V4 Flash 接招，国产模型卡出双线赛道","gpt-5-6-luna-price-cut-equal-intelligence-cost","2026-08-12T03:00:00+00:00"]