[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-intbmoe-full-participation-block-moe":3,"topics-all":38,"news-related-3559e613-9558-48e1-ab20-f53b62796363":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"3559e613-9558-48e1-ab20-f53b62796363","让每个 token 用上全部专家:高德 IntBMoE 解耦参与度、计算与显存,60ms 服务数亿用户","高德团队发布 IntBMoE 论文:用块级条件化让 MoE 的参与度、计算量、参数实例量解耦,专家池全参与配合稀疏执行;论文自述已部署于高德生成式推荐系统,60ms 延迟预算下服务数亿用户,在线 A\u002FB 的 UVCTR 相对提升 2.4%,代码开源于 DreamX-Rec。","MoE(混合专家)架构这两年撑起了几乎所有大模型的容量扩张,但有一个结构性矛盾长期被绕过去:对单个 token 而言,\"多少专家参与贡献\"、\"实际计算多少次专家前向\"、\"要在显存里存多少套专家参数\",这三个量在现有设计里是绑死的。高德团队 9 月发布的 IntBMoE 论文(arXiv:2609.21346)给出一组解耦方案,论文自述已部署在高德的生成式推荐系统里服务数亿用户。\n\n## 三个量为什么一直绑着\n\n论文给三个量下了明确定义:participation(参与度)指一个 token 的输出由多少专家贡献知识;execution(计算量)指实际发生多少次专家计算;materialization(实例化)指需要构建并存储多少套专家级参数。现有三条路线各卡一头:稀疏路由把计算量和实例化压得很低,但每个 token 只有少数专家参与;稠密输出混合恢复了全员参与,计算量却随专家数增长;参数合并把计算压到一个专家,实例化又随路由决策数膨胀。换句话说,过去你最多同时优化两个量,第三个必然恶化。\n\n## IntBMoE:先组合,再稀疏执行\n\nIntBMoE 的核心是\"块级条件化\":从一个小的可学习码本取出块配置,每层由轻量超网络把该层专家池中的全部专家基座合并成组合专家。这样参与度天然拉满——每个组合专家都动用整个池子;执行仍然稀疏——路由器只把 token 送到少数几个块;实例化有上界——块数量由码本决定,与输入无关。作者还设计了双路径残差门控(DPRG),将两条独立组合的路径通过乘法门控耦合,配合特征过滤和一个常驻共享专家完成整体变换。\n\n一个实用细节值得划重点:块可以预先组合并缓存,块配置固定时推理计算量与专家数量无关。这是它区别于稠密混合方案的关键——容量扩张不再直接换算成推理账单。\n\n## 生产成绩单:官方口径\n\n按论文自述,IntBMoE 已完整部署于高德的生成式推荐系统,在 60ms 延迟预算内服务数亿用户,在线 A\u002FB 测试拿到 UVCTR 相对提升 2.4%。实验侧,图像分类任务上稳定超过代表性的稀疏与稠密 MoE 基线,语言建模与序列推荐验证了跨域泛化。需要说明:在线数字全部来自官方 A\u002FB 自报,暂无独立复现,阅读时建议带上这层滤镜。\n\n## 开源仓库给了什么,没给什么\n\n代码落在 GitHub 的 DreamX-Rec 仓库(Apache 2.0,当前 121 stars)。该仓库承载高德六篇生成式推荐工作,IntBMoE 是其中负责容量扩展的模块,配套推荐、自然语言(MiniPile)、视觉(ImageNet-1K)三个域的示例。两点冷静面:其一,开源实现里 BlockMoE 的前向每次调用都重新组合参数,论文研究的 serving 缓存并未包含在公开代码里;其二,推荐示例训练的是单任务 POI 预测,而非完整的多任务目标。在线收益是高德内部系统的结果,公开代码与复现它之间还有明确距离——README 自己也写明了这条边界。\n\n对推理架构方向的人,这篇论文的真正增量不是那 2.4%,而是把 MoE 的成本模型从\"专家数单一旋钮\"细化成三个可独立调节的量。当社区一边苦恼专家并行的通信开销、一边苦恼显存装不下全量专家时,\"先组合、后执行\"提供了第三条路径。下一个值得盯的节点,是社区能否把 serving 缓存补出来——那才是这套设计能否走出高德、进入通用 LLM 栈的胜负手。\n\n参考:论文 arxiv.org\u002Fabs\u002F2609.21346;代码 github.com\u002FAMAP-ML\u002FDreamX-Rec","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.21346","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"d11f0044-8aef-487c-bebe-89ce4683a4a3","moe",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"59174b0b-2d24-4072-b608-d18628a12efd","en","IntBMoE: AMap decouples MoE participation, compute and memory","AMap's IntBMoE decouples MoE participation, execution and materialization via block composition, deployed under a 60ms budget with UVCTR +2.4%.","Mixture-of-Experts has carried most of the capacity scaling in large models over the past two years, but one structural trade-off keeps getting routed around: for a single token, how many experts contribute knowledge, how many expert passes actually get computed, and how many expert-sized parameter sets must be built and stored — in existing designs these three quantities are locked together. AMap's September paper, IntBMoE (arXiv:2609.21346), proposes a decoupling scheme, and the paper states it is already deployed in AMap's generative recommendation system serving hundreds of millions of users.\n\n## Why the three quantities were locked\n\nThe paper defines them precisely. Participation is how many experts contribute knowledge to a token's output; execution is how much expert computation actually runs; materialization is how many expert-sized parameter sets must be built and stored. Three existing routes each get stuck on one: sparse routing keeps execution and materialization low but shrinks participation to a few experts per token; dense output-mixing restores full participation but execution grows with the expert count; parameter-merging keeps execution at one expert while materialization grows with the number of routing decisions. You could optimize at most two of the three — the third always got worse.\n\n## IntBMoE: compose first, execute sparsely\n\nIntBMoE's core is block-level conditioning: block configurations come from a small learned codebook, and at each layer a lightweight hypernetwork merges every expert base in that layer's pool into composed experts. Participation is full by construction — every composed expert draws on the entire pool. Execution stays sparse — a router sends each token to only a few blocks. Materialization is bounded — the codebook, not the input, fixes how many blocks exist. A Dual-Path Residual Gating (DPRG) design couples two independently composed paths through multiplicative gating, completed by feature filtering and an always-active shared expert.\n\nOne practical detail deserves emphasis: blocks can be precomposed and cached, so with a fixed block configuration, inference compute is independent of the expert count. This is what separates it from dense-mixing approaches — capacity scaling no longer translates directly into an inference bill.\n\n## Production numbers, vendor-reported\n\nPer the paper's own account, IntBMoE is fully deployed in AMap's generative recommendation system, serving hundreds of millions of users under a 60ms latency budget, with a 2.4% relative UVCTR gain in online A\u002FB testing. On the experimental side, it shows consistent gains over representative sparse and dense MoE baselines on image classification, with language modeling and sequential recommendation validating generalization across domains. Caveat: the online figures are all vendor-reported from official A\u002FB tests with no independent replication yet — read them with that filter on.\n\n## What the open-source repo gives, and what it doesn't\n\nThe code lives in the DreamX-Rec repository on GitHub (Apache 2.0, currently 121 stars). The repo hosts six AMap generative-recommendation works, with IntBMoE as its capacity-expansion module, shipping recommendation, NLP (MiniPile), and vision (ImageNet-1K) examples. Two sober notes: first, in the open-source implementation, BlockMoE's forward pass recomposes parameters on every call — the serving cache studied in the paper is not part of the public code. Second, the recommendation example trains single-task POI prediction, not the full multi-task objective. The online gains came from AMap's internal system, and there is a clearly marked distance between the public code and reproducing them — the README states this boundary itself.\n\nFor people working on inference architecture, the real increment in this paper is not the 2.4% — it is refining the MoE cost model from a single \"expert count\" knob into three independently tunable quantities. While the community simultaneously struggles with expert-parallel communication overhead and memory that cannot hold the full expert pool, \"compose first, execute sparsely\" offers a third path. The next milestone to watch is whether the community fills in the serving cache — that is the hinge on whether this design escapes AMap into the general LLM stack.\n\nReferences: paper arxiv.org\u002Fabs\u002F2609.21346; code github.com\u002FAMAP-ML\u002FDreamX-Rec","intbmoe-full-participation-block-moe","2026-09-21T13:01:54Z","2026-09-21T13:13:41.451784Z","2026-09-21T13:13:41.451791Z",true,"agent",30,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"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":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"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":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"d941056b-c2e7-42e5-965a-a982c20b1169","Qwen3.8-Flash-Next 架构细节:Gated Residual 多分支残差 + QSA micro-block 稀疏注意力","qwen3-8-flash-next-cost-efficiency-architecture","2026-09-02T02:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"e48b82c0-a106-4a38-b1ff-099cad4db485","四招压平MoE训练显存:Salesforce冲进1M上下文","moe-long-context-memory-peaks","2026-09-17T19:08:39+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"4497a0c5-e9b8-42d5-8d72-cfa0a49c1fba","Mistral Small 4 加入 Firefox Smart Window：开放权重模型第一次进浏览器助手默认菜单","mistral-mozilla-firefox-smart-window-moe","2026-09-17T19:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"f7234b7e-a2c3-404b-9fc4-aaca8e0c8f91","Edge0 预测路由:35B MoE 挤进 24GB Mac","edge0-prerouter-ssd-moe","2026-09-17T15:10:05+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"21fe3c11-4ba4-4801-b6fc-60c4ae559dc1","Yandex 逆流开源:35B 参数的 T5 MoE,每个 token 只激活 0.6B","yandex-aliceai-t5-sparse-moe","2026-09-16T19:11:43+00:00"]