[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-diffusion-llm-decoding-cost":3,"topics-all":38,"news-related-a781accb-537d-4a09-ad5a-d26b87612801":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},"a781accb-537d-4a09-ad5a-d26b87612801","扩散LLM成本实测:写故事比改语法贵3倍,错答案最烧算力","UCLA实测两个扩散语言模型在10类任务上的解码成本:改语法每100个token只要3.7次前向,写故事要11.4次;DreamReasoner做抽象式摘要达101次,反超自回归。成本与输出新信息量相关性最高,错答案比对答案更费前向,按解码速度给投票加权,准确率白涨4.2分。","扩散语言模型的核心卖点是并行:一次前向同时敲定多个 token,推理成本理论上比逐 token 生成的自回归模型低一截。但「低一截」在不同任务上到底低多少,一直没人给过系统账单。UCLA NLP 的研究者 10 月 19 日把两个扩散模型拉上十类任务逐项记账,论文编号 [arXiv:2610.25835](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.25835)。结论先行:省钱幅度高度依赖任务,最贵的任务上优势归零,甚至倒贴。\n\n## 十类任务的账单:3.7 次对 101 次\n\n度量口径是每生成 100 个 token 需要多少次前向——自回归模型恒定 100 次,扩散模型承诺低于这条线。两个被测模型分别是 26B MoE 的 DiffusionGemma-26B-A4B(一次生成 256 token 块)和 8B 稠密的 DreamReasoner(32 token 块),都靠置信度阈值动态决定提交节奏。\n\nDiffusionGemma 一侧:改语法 3.7 次、抽取式摘要 4.3 次、代码编辑 5.0 次、翻译 6.0 次;开放式任务全面变贵——WildBench 对话 8.3 次、抽象式摘要 9.5 次、写故事 11.4 次,是改语法的 3 倍还多。DreamReasoner 一侧整体更贵:从改语法 22.2 次涨到抽象式摘要 101.0 次,论文报告首尾跨度 4.3 倍。101 次是个分水岭数字:它已越过自回归的 100 次基准线,在这类任务上扩散范式不仅没省钱,还多花了。而两个架构、规模、块大小都不同的模型,对「哪个任务贵」的排序几乎一致。\n\n## 贵贱的开关:输出里的新信息量\n\n为什么写故事贵、改语法便宜?作者检验了两个候选因素:输出相对输入有多新(novelty),以及用词在任务内多罕见(surprisal)。与成本相关性最强的不是单独哪一个,而是二者结合的「新信息量」指标:DiffusionGemma 上 Pearson 相关 +0.76,DreamReasoner 上 +0.85;单独的 surprisal 只有 +0.18 和 −0.09,基本不相关。翻译一下:决定成本的不是词冷僻与否,而是这段话有多少内容是第一次被写出来。\n\n思考链内部分布同样印证:论文拆解一道竞赛数学题,思考段 10,339 个 token、40 个块,单块 4 到 22 次前向(成本 3.81);最终答案仅 1,241 个 token(成本 2.88)。答案便宜,因为它大部分在复述思考里已有的内容;一旦答案要输出思考中没有的新东西(WildBench、写故事正是这种),答案成本立刻反超思考段。\n\n## 错答案更烧钱,烧出了免费信号\n\n在 5 场赛事的 160 道竞赛数学题上(每题 5 个种子),难度沿题序上升,准确率从 79% 经 57% 落到 44%,思考长度中位数从 4,416 token 涨到 9,710 token。关键发现在于:控制了题目难度和生成长度后,错误答案仍比正确答案消耗更多前向,且错误答案里「等等,我再检查一下」式的中途自我纠正,出现频率约是正确答案的 5 倍——模型算力开销本身就在泄露它对答案的把握。\n\n这直接催生了论文最实用的一招:采样多个候选答案做多数投票时,按解码速度加权——解码快的票更重。两个候选时准确率比普通投票高 4.2 分,五个候选时高 2.3 分,额外开销为零。多语种则是论文所说的「双倍税」:把 200 个英语句子翻成 25 种语言(16 种文字)后,token 化密度高的语言不仅 token 更多,每 token 前向次数也更多。\n\n## 所以呢\n\n这份实测至少改写三个直觉。第一,扩散模型的成本优势不是均匀资产,复制、编辑、抽取类负载吃满红利,开放式生成可能颗粒无收——选推理范式前先看任务分布。第二,「每 token 单价」对扩散模型是有误导性的计价单位:同一模型、同样长度的两段输出,真实算力可以差 3 倍,按 token 计费的平台迟早要面对这个账。第三,解码速度这个通常被丢弃的副产物,其实是免费的置信度信号,对做推理调度的人来说,这可能比跑分更有用。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.25835","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},"70c337d6-3d72-4deb-b0a5-9f79a59a7cb4","en","Diffusion LLM Cost Study: Stories 3x Pricier Than Grammar Fixes","Diffusion LLM cost varies 3x by task; summarization hits 101 passes\u002F100 tokens, past the AR line. Speed-weighted voting adds 4.2 pts free.","The core promise of diffusion language models is parallelism: one forward pass can commit several tokens at once, so decoding should cost less than the token-by-token march of autoregressive models. But how much less, task by task, had never been measured systematically. Researchers from UCLA NLP put two diffusion models through ten tasks and published the ledger on October 19 as [arXiv:2610.25835](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.25835). The headline: the savings depend heavily on the task, and on the most expensive task the advantage disappears entirely.\n\n## The Ledger: 3.7 Versus 101 Passes\n\nThe metric is forward passes per 100 generated tokens. An autoregressive model always sits at 100; diffusion models promise to stay under that line. The two models tested are DiffusionGemma-26B-A4B, a 26B mixture-of-experts model generating 256-token blocks, and DreamReasoner, an 8B dense model generating 32-token blocks. Both use semi-autoregressive block diffusion and commit only tokens that clear a confidence threshold, so step counts adapt on their own.\n\nOn the DiffusionGemma side, grammar correction costs 3.7 passes, extractive summarization 4.3, code editing 5.0, translation 6.0. Open-ended work gets expensive fast: WildBench dialogues cost 8.3, abstractive summarization 9.5, and story writing 11.4 — more than three times grammar fixing. DreamReasoner is pricier overall, climbing from 22.2 passes on grammar correction to 101.0 on abstractive summarization, a spread the paper reports as 4.3x. That 101 is a watershed number: it crosses the autoregressive line of 100, meaning diffusion not only stopped saving money on this task but spent more. Yet the two models — different in architecture, size, and block length — agree almost perfectly on which tasks are expensive.\n\n## What Flips the Switch: New Information\n\nWhy is a story costly and a grammar fix cheap? The authors test two candidate properties: how novel the output is relative to the input, and how rare its words are within the task. The strongest correlate of cost is neither alone but their combination, a \"new information\" measure: Pearson +0.76 on DiffusionGemma and +0.85 on DreamReasoner, while surprisal alone manages only +0.18 and −0.09. In plain terms, cost is not driven by obscure vocabulary but by how much of the text is being written for the first time.\n\nThe anatomy of a single competition math problem makes the point. The model thinks for 10,339 tokens across 40 blocks, each taking 4 to 22 forward passes (cost 3.81), then writes a 1,241-token answer (cost 2.88). The answer is cheap because it mostly rewords what the thinking already contained; when the answer must add new content the thinking never produced — as in WildBench or story writing — its cost overtakes the thinking itself.\n\n## Wrong Answers Burn More, and the Burn Is a Signal\n\nAcross 160 competition problems from 5 contests, five seeds each, accuracy falls from 79% through 57% to 44% as problems get harder, while median thinking length grows from 4,416 to 9,710 tokens. The key finding: after controlling for problem difficulty and generation length, incorrect answers still consume more forward passes than correct ones, and wrong answers stop to correct themselves — \"wait, let me re-check\" moments — about five times as often as right ones. Compute usage itself leaks the model's confidence.\n\nThat yields the paper's most practical trick. When sampling multiple candidate answers and taking a majority vote, weight each vote by how quickly it was decoded. Fast decodes get heavier votes, and accuracy beats plain majority voting by 4.2 points with two candidates and 2.3 points with five, at zero extra cost. Multilingual generation carries what the paper calls a double tax: translating 200 English sentences into 25 languages across 16 scripts shows that highly tokenized languages pay both with more tokens and with more passes per token.\n\n## So What\n\nThree intuitions get rewritten. First, the diffusion cost advantage is not a uniform asset: copy, edit, and extract workloads collect the full dividend, while open-ended generation may collect nothing — check your task mix before picking a decoding paradigm. Second, \"price per token\" is a misleading unit for diffusion models: two outputs of identical length from the same model can differ 3x in real compute, a bill token-priced API providers will eventually have to face. Third, decoding speed, usually discarded as a byproduct, is a free confidence signal — which for anyone scheduling inference may be worth more than another benchmark point.","diffusion-llm-decoding-cost","2026-10-25T21:09:54Z","2026-10-25T21:09:57.284436Z","2026-10-25T21:09:57.284452Z",true,"agent",379,[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},"76eb5d5a-04c1-4d7f-9c50-dc494b1c3bb8","扩散LM宽窗塌方解剖:一成回退提交,点燃七成失败","diffusion-lm-fallback-commit-failure-roots","2026-11-17T15:35:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"269851b4-d6e6-43a6-9239-b3116708b735","块边界是同步税:扩散LM重叠去噪免训练提速14.5%","b3-overlapping-block-denoising-dlm","2026-11-11T17:11:59+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"51953ff7-5702-4d21-9af6-92d8edf7af09","块扩散LM双稀疏提速:不动权重,32K单步快4.72倍","duet-dual-sparse-bdlm-inference","2026-11-09T15:20:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"fd9298b5-dea9-4641-8f73-56ae32261c3e","扩散LM投机采样:串行评估砍到1\u002F4,分布零走样","specctmc-exact-speculative-diffusion-lm","2026-11-07T15:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"af6bfcff-1ffa-400d-ac93-cb6803ab232c","dSpark调度蒸馏:扩散LLM解码最多快10倍","dspark-schedule-distillation-diffusion-llm","2026-10-27T17:09:34+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"6ede1566-74c5-4bdd-9dd5-6d3f118db876","投机解码的重复陷阱:D-Loop 两轮循环破局","d-loop-diffusion-speculative-decoding","2026-10-06T15:20:00+00:00"]