[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-d-loop-diffusion-speculative-decoding":3,"topics-all":38,"news-related-6ede1566-74c5-4bdd-9dd5-6d3f118db876":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},"6ede1566-74c5-4bdd-9dd5-6d3f118db876","投机解码的重复陷阱:D-Loop 两轮循环破局","港城大、港中文与华为的 D-Loop 论文发现块扩散草稿的「重复陷阱」:相邻位置各自预测互不通气,MT-Bench 上 42.6% 草稿块相邻重复,拖短接受长度。D-Loop 不加新模块,让草稿器循环跑两遍,第二遍按前缀重写后缀,Qwen3-4B 八基准平均加速 5.06 倍,采样时优势扩大到 26.2%。","投机解码的通行做法是让小模型打草稿、大模型并行验收。块扩散把这件事推进一步:一个轻量扩散草稿器在前向里一次吐出一整块 token,而不是逐个猜。但香港城市大学、香港中文大学与华为研究团队在 10 月 5 日放出的 D-Loop 论文里,指出了这套方案的一个系统性毛病——他们称之为「重复陷阱」:块内每个位置都在预测自己的边缘分布,谁也看不见谁,结果相邻位置常常连发同一个 token。草稿一旦重复,验收时就会被目标模型拦腰截断,接受长度上不去,加速比自然缩水。\n\n## 重复陷阱有多严重\n\n论文统计了 DFlash 草稿输出里的重复情况(接受长度按 Qwen3-8B、采样解码计)。GSM8K 上相邻重复率 17.4%,LiveCodeBench 29.4%,到对话基准 MT-Bench 已经到 42.6%;连续重复内容词出现在 72.5%–94.5% 的草稿块里,MT-Bench 上最长的连续重复段平均 4.7 个 token。而三个基准对应的接受长度分别是 5.28、3.26 和 3.10——重复越凶,接受越短。\n\n## 两遍循环,一个模型\n\n此前的修法是给草稿器外挂一个因果头,或者再训一个独立草稿器,参数和训练目标都变多。D-Loop 的思路相反:不加任何新模块,让同一个草稿骨干循环跑两遍。第一遍照常提议一整块;第二遍选定一个前缀作为条件,并行重写剩下的后缀。配套的「前缀–后缀互补目标」让同一组参数同时学会「无锚点预测前缀」和「给定前缀预测后缀」两种模式。消融显示,仅第二轮(K=1)就把 GSM8K、LiveCodeBench、MT-Bench 的相邻重复率砍掉 73.7%、83.0% 和 79.8%。\n\n## 加速数字\n\n在数学、代码、对话共八个基准上(实测 GSM8K、MBPP、LiveCodeBench、MT-Bench、Alpaca 等),Qwen3-4B 在两种温度下的接受长度都在作者实测的方法里排在第一位,平均加速 5.06×(贪心)与 4.11×(采样);Qwen3-8B 上平均接受长度 7.19 与 5.77,平均加速 5.02× 与 4.03×。值得留意的是采样场景:相对 DSpark 的平均加速优势在 4B 上从 2.2% 扩大到 15.4%,在 8B 上从 9.7% 扩大到 26.2%——温度一高,边缘分布的重复病更重,第二遍循环的价值也就更大。\n\n## 所以呢\n\n对部署侧来说,采样解码才是对话与创意生成的日常状态,D-Loop 的增益恰好在这种场景放大,而且不加参数、不引入第二套训练目标,工程上好接。当然,它对比的 DFlash 与 DSpark 是作者自己复现的版本,EAGLE-3 数字引自 DFlash 论文(tree size 60),不同树大小下的对比还留有讨论空间。但「用循环换质量、用共享参数换部署成本」这条路线,确实给扩散式投机解码补上了一块关键拼图——详见 [arXiv:2610.06011](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.06011)。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.06011","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"d8d7f86e-34de-44ac-a816-97bb23bfcd9c","en","D-Loop Breaks the Repetition Trap in Speculative Decoding","CityU HK, CUHK and Huawei find block-diffusion drafts fall into a repetition trap. D-Loop loops the drafter twice and hits 5.06x speedup on Qwen3-4B.","Speculative decoding usually lets a small model draft and a large model verify in parallel. Block diffusion pushes this further: a lightweight diffusion drafter emits an entire block of tokens in one forward pass instead of guessing one by one. But the D-Loop paper, released on October 5 by City University of Hong Kong, The Chinese University of Hong Kong and Huawei Research, identifies a systematic flaw they call the \"repetition trap\": every position inside the block predicts its own marginal distribution without seeing its neighbors, so adjacent positions often emit the same token over and over. Once a draft repeats, verification cuts it short, the acceptance length stalls, and the speedup shrinks.\n\n## How bad is the repetition trap\n\nThe paper measures repetition in raw DFlash drafts (acceptance lengths on Qwen3-8B, sampling). The adjacent-repeat rate is 17.4% on GSM8K, 29.4% on LiveCodeBench, and 42.6% on MT-Bench; consecutive repeated content words appear in 72.5%–94.5% of draft blocks, and the longest repeated run on MT-Bench averages 4.7 tokens. The corresponding acceptance lengths are 5.28, 3.26 and 3.10 — heavier repetition, shorter acceptance.\n\n## Two looped passes, one model\n\nEarlier fixes bolt a causal head onto the drafter or train a separate drafter, adding parameters and training objectives. D-Loop does the opposite: no new modules, just the same drafting backbone run twice. The first pass proposes a full block; the second conditions on a selected prefix and regenerates the suffix in parallel. A complementary prefix–suffix objective trains the shared parameters for both anchor-only prefix prediction and prefix-conditioned suffix prediction. With only the second loop (K=1), adjacent-repeat rates drop by 73.7%, 83.0% and 79.8% on GSM8K, LiveCodeBench and MT-Bench.\n\n## The speedup numbers\n\nAcross eight math, code and chat benchmarks (including GSM8K, MBPP, LiveCodeBench, MT-Bench and Alpaca), D-Loop achieves the longest acceptance length among the methods the authors ran themselves on Qwen3-4B at both temperatures, with average speedups of 5.06x (greedy) and 4.11x (sampling); on Qwen3-8B it reaches average acceptance lengths of 7.19 and 5.77 with average speedups of 5.02x and 4.03x. The sampling regime matters most: the average-speedup edge over DSpark widens from 2.2% to 15.4% on the 4B model and from 9.7% to 26.2% on the 8B — higher temperature worsens the marginal-distribution repetition disease, which makes the second loop more valuable.\n\n## So what\n\nFor deployment teams, sampling is the everyday mode for chat and creative generation, and that is exactly where D-Loop's gains grow — with no extra parameters and no second training objective, which keeps integration cheap. Caveats: DFlash and DSpark comparisons come from the authors' own runs, and EAGLE-3 numbers are quoted from the DFlash paper (tree size 60), so other tree sizes remain open for discussion. Still, the route of \"trading loops for quality and shared parameters for deployment cost\" adds a key piece to diffusion-style speculative decoding — see [arXiv:2610.06011](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.06011).","d-loop-diffusion-speculative-decoding","2026-10-06T15:20:00Z","2026-10-06T15:13:21.883582Z","2026-10-06T15:13:21.883596Z",true,"agent",675,[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},"8b50b8ec-0249-420f-a091-0d8b7efe6a18","Flash-dLLM 免训加速扩散语言模型:最高148倍","flash-dllm-io-aware-kv-cache","2026-09-23T15:10:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"8914cfad-53bc-4f76-bb06-8579d23cfa5d","AR 大模型外挂 diffusion 权重:Uno 免草稿模型实现 3 倍无损解码加速","uno-diffusion-augmented-llm-speedup","2026-09-08T13:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"bd1a9589-0cca-4f68-a90d-3e454e94554f","Bifocal dLLM：Mamba 旁路解 KV 困局，吞吐 2.4×–12.9×","bifocal-dllm-r2lm-mamba-qwen3-1-7b","2026-06-29T10:08:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"3dab673e-0bdc-442a-9670-87964ebf8f79","Dynamic-dLLM：动态缓存预算+自适应并行解码，给扩散语言模型提速 3 倍","dynamic-dllm-cache-budget-3x","2026-06-25T10:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"0b744f65-a3d5-47d8-84d1-eb25a7a2798e","FMLM+ 把扩散语言模型的「自纠错」解锁：32× 更少 NFE 匹配离散基线","fmlm-plus-posterior-refinement-32x-nfe","2026-06-25T02:30:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"80bc5e25-d24d-45a1-9c2b-534ebcae39f9","腾讯 WeDLM 开源：让扩散 LLM 在标准因果注意力下跑出 3-6× vLLM 加速","tencent-wedlm-diffusion-llm-causal-3-6x","2026-06-16T20:00:00+00:00"]