[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-baidu-ernie-5-1-elastic-moe-6pct-cost":3,"topics-all":36,"news-related-937e2875-f1ac-42f5-8b0d-85733ca7e064":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},"937e2875-f1ac-42f5-8b0d-85733ca7e064","百度文心5.1发布：预训练成本仅占业界6%，弹性蒸馏技术解析","百度正式发布文心大模型5.1，基于多维弹性预训练技术，仅以业界同规模模型约6%的预训练成本，达到基础效果领先水平，并登上LMArena搜索榜国内第一、全球第四。\n\n这项技术的核心是Once-for-All弹性训练框架，通过弹性深度（随机跳过Transformer层）、弹性宽度（动态调控MoE专家池）、弹性稀疏度（可变Top-k路由）三个维度，实现了一次预训练生成多种规模模型的能力，将总参数量压缩至文心5.0的约1\u002F3，激活参数压缩至约1\u002F2。\n\n后训练阶段采用分离式全异步强化学习架构，将训练、推理、奖励、智能体循环四大子系统控制面完全解耦，配合FP8低精度算子库和OPD多阶段训练管线，有效解决了跷跷板效应问题，实现了Agent、推理、创作等多维能力的均衡提升。在AIME26数学竞赛评测中，文心5.1得分99.6，仅次于Gemini 3.1 Pro。\n\n文心5.1的意义不仅在于单次训练的效率突破，更在于弹性训练范式本身——用更少资源做更多模型，正在成为大模型竞争的新分水岭。","https:\u002F\u002Fyiyan.baidu.com\u002Fblog\u002Fzh\u002Fposts\u002Fernie-5.1-0508-release\u002F","b51e30de-7256-42c6-a5bd-409037e1f9e7",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",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},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"2f106e28-e16d-4ab1-bb8d-accacbedfca6","en","Baidu ERNIE 5.1: pretraining at 6% of industry cost","Baidu has officially released ERNIE LLM 5.1, based on multi-dimensional elastic pretraining technology, achieving leading baseline performance with only about 6% of the pretraining cost of same-scale industry models, and topping LMAarena's search ranking #1 in China and #4 globally.\n\nThe core of this technology is the Once-for-All elastic training framework. Through three dimensions — elastic depth (randomly skipping Transformer layers), elastic width (dynamically adjusting the MoE expert pool), and elastic sparsity (variable Top-k routing) — the framework achieves the ability to generate models of various sizes from a single pretraining run, compressing total parameters to about 1\u002F3 of ERNIE 5.0, and activated parameters to about 1\u002F2.\n\nThe post-training stage adopts a decoupled fully-asynchronous RL architecture, with the control planes of training, inference, reward, and agent loop subsystems fully decoupled, combined with an FP8 low-precision operator library and OPD multi-stage training pipeline, effectively solving the seesaw-effect problem, and achieving balanced improvement across Agent, reasoning, and creative capabilities. On the AIME26 math competition evaluation, ERNIE 5.1 scored 99.6, second only to Gemini 3.1 Pro.\n\nThe significance of ERNIE 5.1 goes beyond a single training efficiency breakthrough — it lies in the elastic training paradigm itself: using fewer resources to build more models is becoming the new dividing line in LLM competition.","baidu-ernie-5-1-elastic-moe-6pct-cost","2026-05-09T10:10:00Z","2026-05-09T10:05:38.481144Z","2026-08-19T02:08:40.142862Z",true,"agent",236,[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},"2b37a19b-1dde-4238-bef5-39b1d19157f1","OpenBMB 开源 MiniCPM5-2B:2B 端侧模型平均分超对比集 4B 级","openbmb-minicpm5-2b-on-device","2026-09-07T17:02:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"8c6d408b-2913-48b2-bc1f-4e958ad854c5","Space Bunny 谜底揭晓:MiniMax M3.1 Flash,匿名期吞下 200 万亿 token","minimax-m3-1-flash-space-bunny","2026-10-19T21:10:12+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"e304fe91-16d5-4c28-81c7-0c4c0dd8c669","小米公开MiMo-V2.6训练账本:RL烧了260万美元","xiaomi-mimo-v2-6-rl-scaling-ledger","2026-10-10T13:11:49+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"890ad5c1-b8de-4df3-ab96-67dceca82841","阿里千问下一代冲到 5-10 万亿参数","alibaba-qwen-next-gen-5-10-trillion-parameters","2026-09-29T02:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"eac420e1-d05e-457e-a059-b4d724d36620","阿里云栖大会:Qwen 路线图拉到 5-10 万亿参数,真武 V900 性能三倍","alibaba-qwen-5-trillion-zhenwu-v900","2026-09-22T07:30:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"2bd44b6f-5688-471f-930e-17a93984e8e7","中国电信开源星辰 Xing4.0:昇腾全栈训练的 29B MoE","xing4-29b-a4b-ascend-moe","2026-09-19T15:10:00+00:00"]