[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-d-cut-speculative-draft-cut":3,"topics-all":36,"news-related-8a42c9c3-a1c7-40fb-8c75-8ac42977b5af":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},"8a42c9c3-a1c7-40fb-8c75-8ac42977b5af","D-cut 把投机解码的「长草稿」剪掉一半：高并发推理平均提速 1.65×、MoE 跑出 3×","投机解码（speculative decoding）一直被认为是「无损加速 LLM 推理」的标准配方，但 D-cut（arXiv:2607.14647）把它从教科书推到了真实部署现场：在高并发批处理场景下，长 draft 反而让验证过程消耗大量算力、把投机解码拖得比自回归还慢，这正是当前 vLLM、TGI、TensorRT-LLM 等推理引擎内部最棘手的尾部开销。D-cut 的关键设计只有两条：第一是跨请求联合剪枝（cross-request pruning），把同一个 mini-batch 里所有正在跑的请求当成一张「草稿接受长度热力图」，按草稿置信度重新分配验证预算，让高接受率的请求多验、低接受率的请求少验，避免为注定被拒的 token 白算；第二是把剪枝深度和目标硬件绑定——GPU 架构、并行策略、张量并行度都会改写「验证一个 token」的代价，D-cut 内置一份 runtime cost model，让深度自动适配到 H100、B200 还是消费级卡。实际效果：在 dense 与 MoE 模型上，并发打满时平均 speedup 由 1.26× 跳到 1.65×；部分 dense 配置下，原投机基线已经输给自回归，D-cut 把加速「救回来」；MoE 场景最高 3.0× 速度，逼近单请求理想态。这是一篇工程味很浓的论文，没有改模型结构，不追求榜单虚名，而是把推理服务里那 10%-30% 的尾部时延稳吃下来。值得指出的是，近两年推理优化已经走过「KV cache 量化、推测解码、稀疏 attention」三波，D-cut 把焦点放在 batch 维度的 budget allocation，意味着 LLM serving 的下一战场不再是单请求极致加速，而是「多请求协同调度」的系统题。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.14647","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"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},"045c011e-e2bb-45ce-bdd6-0c927f8a3b87","token-efficiency",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"fd3aa531-330b-4fd9-8891-88db64342767","en","D-cut halves speculation's long drafts, 1.65x to 3x speedups","Speculative decoding has long been treated as the standard recipe for \"lossless LLM inference acceleration\", but D-cut (arXiv:2607.14647) pushes it from textbook to real production: under high-concurrency batch scenarios, long drafts cause the verification process to burn a lot of compute, dragging speculative decoding slower than autoregression — this is exactly the tail-latency pain inside vLLM, TGI, TensorRT-LLM, and other inference engines. D-cut's design has only two pillars: first, cross-request pruning — turning all currently-running requests in the same mini-batch into a \"draft-acceptance-length heatmap\", reallocating the verification budget by draft confidence, so high-acceptance requests verify more, low-acceptance ones less, avoiding burning tokens on those doomed to be rejected; second, tying pruning depth to the target hardware — GPU architecture, parallel strategy, tensor-parallel degree all rewrite the cost of \"verifying one token\", and D-cut embeds a runtime cost model that automatically adapts depth to H100, B200, or consumer-grade cards. The actual effect: on dense and MoE models, average speedup with concurrency maxed jumps from 1.26× to 1.65×; in some dense configurations the original speculative baseline already loses to autoregression, and D-cut \"saves back\" the speedup; MoE scenarios hit up to 3.0× speed, approaching the single-request ideal. This is a strongly engineering-flavored paper, no architectural changes, no leaderboard flexing, it just steadily eats the 10%-30% tail latency in inference serving. Worth pointing out: in the past two years inference optimization has gone through three waves — \"KV cache quantization, speculative decoding, sparse attention\" — and D-cut's focus on batch-dimension budget allocation means the next battlefield for LLM serving is no longer single-request extreme acceleration, but a systems problem of \"multi-request cooperative scheduling\".","d-cut-speculative-draft-cut","2026-07-18T10:10:00Z","2026-07-18T10:14:09.478652Z","2026-08-19T02:08:40.142862Z",true,"agent",271,[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},"3096df88-7158-4ffe-9356-1a83b829633b","A*-Thought-V2:把思维链塞进隐空间,回复砍半,平均精度反升","astar-thought-v2-latent-cot-compression","2026-09-09T15:10:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"5c75bd80-9f12-499a-898f-019615ac98ee","Prefix Sliding:让推理模型长思考提速3倍的免训练方案","prefix-sliding-efficient-test-time-scaling","2026-08-27T17:20:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"21be10ba-3364-4b55-a248-e6f878dae68b","QuasiMoTTo：quasi-Monte Carlo 进 test-time scaling","quasimotto-quasi-monte-carlo","2026-07-02T10:20:52+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"a453eb28-7fb0-4e07-adc1-0d0575850758","EntMTP 用熵信号给多 token 推测装上调速器：让 LLM 自适应匹配上下文可预测性","entmtp-entropy-speculative-decoding","2026-06-29T12:21:51+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"9a3cd449-e29a-4730-814b-f1be5c2685c6","复旦FFD让Flash Attention退役？11.6× kernel提速把长上下文推到256K","fudan-ffd-long-context-attention-sparsity","2026-09-15T07:15:46+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"2638aeac-dc4d-4b73-b7fe-2b042015adee","OreoLook 开源:三层缓存把 AI 搜索搬进 8 核 CPU,重复问题 0.1 毫秒出答案","oreolook-three-layer-cpu-cache","2026-09-10T23:08:36+00:00"]