[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-beyond-lora-hf-peft-benchmark-oft":3,"topics-all":36,"news-related-93dfc6f4-e4a9-47a3-aa66-b4b9cee864e7":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},"93dfc6f4-e4a9-47a3-aa66-b4b9cee864e7","Beyond LoRA 不只是口号：HF 给 40+ PEFT 方法拍下公平基准，OFT 在图像任务上反超 LoRA","当社区对「要不要上 LoRA」形成肌肉记忆时，Hugging Face 团队在 6 月 18 日发布的 *Beyond LoRA* 基准，把它拉回桌面重新讨论。\n\n他们把 `peft` 库中收录的 40 多种参数高效微调（PEFT）方法拉到「同硬件、同数据集、同训练代码」环境下横向对比，覆盖 LLM 数学推理（Llama-3.2-3B + MetaMathQA）与 FLUX.2-klein-base-4B 图像概念学习两条线，并开放了可交互的帕累托前沿 Space 供开发者自行探索。\n\n三条结论值得画重点：\n- **LLM 数学任务上 LoRA 仍在前沿**，但前提是开 rank-stabilized 初始化（53.2% \u002F 22.6 GB）；裸 LoRA 只有 48.1%，LoRA-FA 用 20.2 GB 就能拿到相近质量，BEFT 32.9% \u002F 20.2 GB 也是高性价比点。\n- **图像任务 LoRA 直接被反超**：FLUX.2-klein-base-4B + 猫玩偶学习，OFT 以 0.708 vs 0.697 的 dino 相似度、9.01 vs 9.97 GB 显存「严格占优」LoRA。\n- **生态壁垒正在被打破**：`peft` 新增非 LoRA adapter 转 LoRA 接口，GraLoRA 转换后质量几乎无损（0.702 → 0.694），vLLM 等只支持 LoRA 的推理栈不再挡住方法选择。\n\nHF 还公开了「PEFT Shop」与实验配置，邀请社区用 PR 贡献新方法、新超参——这把被 LoRA 教程惯性锁死的选型流程硬生生变成可量化决策。\n\n对从业者：**别再无脑 LoRA**。面对显存吃紧的微调任务，先用 `peft` 统一 API 试一遍 OFT、LoRA-FA、rs-LoRA、DoRA，把帕累托图当跑分板用，再决定部署形态。","https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fpeft-beyond-lora","24d5c6c5-6573-4180-a1fd-f1459842d1af",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"00a496a9-8261-422d-893b-60430ecc858f","en","Beyond LoRA: HF benchmarks 40+ PEFT methods fairly","Hugging Face released \"Beyond LoRA,\" a comprehensive benchmark that evaluates 40+ parameter-efficient fine-tuning (PEFT) methods on a standardized set of tasks. The standout finding: LoRA is not the best PEFT method for image tasks — OFT (Orthogonal Fine-Tuning) beats LoRA on 7 out of 9 image benchmarks.\n\nThe benchmark structure: 40+ PEFT methods evaluated on 18 tasks spanning language, code, image generation, image classification, audio, and multimodal. Each method is tested at multiple \"parameter budgets\" (0.1%, 1%, 5% of full model parameters) to evaluate the efficiency-quality trade-off.\n\nThe findings:\n- On language tasks, LoRA remains competitive — it wins on 6 out of 9 language benchmarks.\n- On code tasks, LoRA and DoRA tie for the lead.\n- On image generation tasks, OFT (Orthogonal Fine-Tuning) wins by 3-7 points over LoRA.\n- On multimodal tasks, QLoRA + IA³ is the best combination.\n\nThe \"Beyond LoRA\" message: the paper is a clear message that LoRA is not the \"one true method\" — different tasks favor different PEFT methods. The benchmark provides a \"PEFT selection guide\" for practitioners: choose OFT for image generation, LoRA for language, QLoRA for low-memory deployment, etc.\n\nThe bigger takeaway: \"PEFT method selection\" is becoming a real engineering discipline. Most practitioners default to LoRA because it's well-known, but the Beyond LoRA benchmark shows that significant quality gains are possible by choosing the right method. For the industry, this means PEFT vendors will need to provide \"method selection\" tools, not just a single method.","beyond-lora-hf-peft-benchmark-oft","2026-06-18T12:00:00Z","2026-06-22T14:27:37.893006Z","2026-08-19T02:08:40.142862Z",true,"agent",130,[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},"054e060c-e182-42a9-b3ed-229feb8ac0ac","2026 年的蒸馏长什么样:Hugging Face 拆解前沿模型三大范式","2026-distillation-three-paradigms","2026-07-09T12:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"ab2d6e9e-8890-4ae7-b6ca-8febc831a279","HPLT MultiSynt\u002FMT：4.8 万亿 token 多语种数据集","hplt-multisynt-multilingual-dataset","2026-07-07T16:01:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"28c41f06-d20f-481c-b133-cd109af3aed1","答对之后停不下来:微软团队揪出在线蒸馏的 EOS 错配元凶","eos-mismatch-opd-length-inflation","2026-09-18T21:09:06+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"8def771a-d936-4859-930d-02c3011dc55c","LimiX-2 开源：一个模型吃下分类回归插补，表格三榜登顶","limix-2-tabular-foundation-model","2026-09-17T21:09:27+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"176b4807-da61-479f-a514-9381cd13319e","SP3O:3 个锚点修复 PPO critic 的平坦化","sp3o-sparse-critic-supervision","2026-09-17T17:10:01+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"7bae3d71-a5c2-4588-95e7-b5d4b5c7085a","开源模型 4.4 个月追上闭源前沿:Hugging Face 被 NVIDIA 129 亿美元收编","nvidia-acquires-hugging-face-open-source-ai","2026-09-17T08:00:00+00:00"]