[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-speculative-forcing-video-denoising":3,"topics-all":38,"news-related-4349edac-997e-4980-95a0-58054a28a9dd":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},"4349edac-997e-4980-95a0-58054a28a9dd","视频生成投机去噪:1.7 步顶 4 步,提速 1.5 倍","KAIST×NAVER 提出 Speculative Forcing:自回归视频生成引入草稿-验证-提交机制,token 收敛即冻结、不再去噪,不改权重。LongLive 上约 1.7 步保住 4 步质量,H100 提速超 1.5 倍;世界生成场景约 2.5 步,提速 1.35 倍。","流式视频生成走到 few-step 自回归这一代,已能一边生成一边播放。但 KAIST 与 NAVER AI Lab 的团队指出一个推理循环里的浪费:few-step 自回归视频生成器把每个 token 都送满完整去噪 schedule,不管它还需不需要精炼。均匀计算是默认选项,却常无必要——很多 token 在最初几步就非常接近最终输出,另一些则要 refine 到最后。\n\n## 不是每个 token 都值得跑满四步\n\nLongLive 上的步数-质量阶梯很直接:1 步 VBench 总分 79.06,2 步 82.46,3 步 83.08,4 步 83.08。3 到 4 步总分一分不涨;但 1 步掉到 79.06,无脑减步也不可行。项目页可视化显示:背景、与历史帧相似的区域很早收敛,镜头新揭示的区域才需持续去噪。收敛是 token 级事件,不是整段视频的集体属性,这是整套方法的出发点。\n\n## 草稿、验证、提交\n\nSpeculative Forcing 拆成三段。**Draft**:每个去噪步本来就会给出「干净估计」,把它当作该 token 最终输出的草稿。**Verify**:验证器判断草稿距 full-step 输出还有多远——training-free 版基于与干净内容的相似度和跨步稳定性;learned 版是读冻结生成器中间特征的小型 3D 卷积头。**Commit**:通过验证的 token 立即冻结为最终输出,跳过剩余全部去噪步,但其 KV 缓存保留,后续 token 仍在完整上下文里精炼。全程不改生成器权重;推理期只用一个容差 ε 调速度与质量,ε 在 0.20-0.24 区间能保住四步质量且 Drift 最低。\n\n## 三个场景,三组数字\n\n- **5 秒短视频(LongLive)**:learned 验证器平均 1.68 步\u002Ftoken,H100 上 41.74 fps,VBench 总分 83.21;四步基线 4.00 步、26.40 fps、83.08,质量不降反升,吞吐 1.58 倍。training-free 版 1.65 步、43.61 fps、83.12。\n- **30 秒长视频**:1.72 步\u002Ftoken,24.5→37.8 fps(1.54 倍),VBench-Long 总分 82.69→82.81,Drift 从 1.722 降到 1.707,提前冻结反而更稳。\n- **世界生成(Lingbot-World 2.0)**:镜头控制持续揭示新内容,平均步数升到 2.55,仍达 18.1→24.5 fps(1.35 倍),Drift 3.094→2.839。\n\n## 边界与启发\n\n消融给出两个失败模式:粒度从 token 放粗到 chunk,整块过早提交会让误差传到后续 chunk;时间差分当接受信号,分不清镜头运动和生成失真。收益建立在「粒度对、信号对」之上;2.55 步的世界生成也划出边界——新内容越多,可省的计算越少。\n\n这套 Draft-Verify-Commit 拆法,与 LLM 推理里「草稿先行、验证把关」的资源分配逻辑同构:都赌大多数位置不需要最贵的计算。区别在视频的「接受」语义更微妙——冻结的是空间区域的视觉状态,还得为后续 token 留着上下文。\n\n对做流式视频生成与世界模型 rollout 的团队:不改权重的推理期加速,已部署的 few-step 生成器可直接套用;ε 一个旋钮调速度-质量档位。当视频生成向实时交互收敛,每 token 计算量分配会像 LLM 推理优化一样成为独立战场。\n\n参考:arXiv:2610.22347|speculative-forcing.github.io|github.com\u002Fnaver-ai\u002Fspeculative-forcing","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22347","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},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"758c1ba5-441a-4035-b289-17d9f93d703f","en","Speculative Forcing: Video Tokens Freeze Early, 1.5x Faster","Speculative Forcing freezes converged video tokens early: 4-step quality at ~1.7 steps per token on LongLive, over 1.5x faster on one H100.","Streaming video generation has reached the few-step autoregressive generation: models can now generate and play back at the same time. But researchers from KAIST and NAVER AI Lab point out a waste hiding inside the inference loop: few-step autoregressive video generators run every single token through the full denoising schedule, regardless of how much refinement it actually needs. Uniform computation is the default, yet it is often unnecessary — many tokens already closely match their final outputs within the first steps, while others keep changing until the end.\n\n## Not every token deserves four full steps\n\nThe step-count-versus-quality ladder on LongLive (5-second videos) makes the point bluntly: 1 step scores 79.06 on VBench total, 2 steps 82.46, 3 steps 83.08, 4 steps 83.08. Going from 3 to 4 steps adds exactly zero to the total, while 1 step drops to 79.06 — so blindly cutting steps does not work either. The project page visualization goes further: background regions and areas similar to history converge early, while newly revealed regions keep needing denoising. Convergence is a per-token event, not a property of the whole clip — that is the starting point of the whole method.\n\n## Draft, verify, commit\n\nSpeculative Forcing splits into three stages. **Draft**: every denoising step already predicts a clean estimate for each token; treat that intermediate estimate as a draft of the token's final output. **Verify**: a verifier predicts how far each draft still is from the full-step output. The team built two designs — a training-free one based on similarity to clean content and stability across steps, and a learned one, a small 3D convolutional head reading the frozen generator's intermediate features. **Commit**: accepted tokens are immediately frozen as final outputs and skip all remaining denoising steps, while their cached keys and values stay available, so the rest of the tokens keep refining with full context. The generator's weights are never touched; at inference time a single tolerance ε trades speed against quality, with ε between 0.20 and 0.24 preserving four-step quality at the lowest Drift.\n\n## Three scenarios, three sets of numbers\n\n- **5-second short video (LongLive)**: the learned verifier averages 1.68 denoising steps per token, 41.74 fps on a single H100, VBench total 83.21 — versus the four-step baseline at 4.00 steps, 26.40 fps, 83.08. Quality edges up by 0.13 while throughput rises 1.58x; the training-free variant runs 1.65 steps at 43.61 fps with 83.12.\n- **30-second long video**: 1.72 steps per token, 24.5 to 37.8 fps (1.54x), VBench-Long total 82.69 to 82.81, and Drift — the variation of imaging quality over the rollout — drops from 1.722 to 1.707. Freezing early does not hurt consistency; it actually steadies it.\n- **World generation (Lingbot-World 2.0)**: camera control keeps revealing new content, so more tokens need the full schedule. The average climbs to 2.55 steps, yet throughput still goes 18.1 to 24.5 fps (1.35x), with Drift improving from 3.094 to 2.839.\n\n## Design details and the method's boundary\n\nThe ablation study carries two lessons. First, coarsening acceptance granularity from tokens to chunks commits whole chunks too early, and the errors propagate into later chunks. Second, using temporal difference as the acceptance signal cannot tell camera-induced changes from generation distortions. Together these failure modes show that the gains rest entirely on choosing the right granularity and the right signal. The 2.55-step world-generation result also draws the boundary: the more new content in the frame, the less computation there is to save.\n\nIn my view, this Draft-Verify-Commit decomposition is structurally the same bet as draft-then-verify acceleration in LLM inference: most positions do not need the most expensive computation. The difference is that \"acceptance\" in video is subtler — what gets frozen is the visual state of a spatial region, and the context must stay alive for later tokens.\n\nFor teams building streaming video generation and world-model rollouts, the practical meaning is direct: weight-free inference acceleration means deployed few-step generators can adopt it as-is, and the single ε knob dials the speed-quality point per scenario. As video generation converges toward real-time interaction, per-token compute allocation is becoming its own battlefield, much like LLM inference optimization. The code is open-sourced on GitHub.\n\nReference: arXiv:2610.22347 (arxiv.org\u002Fabs\u002F2610.22347) | Project page speculative-forcing.github.io | Code github.com\u002Fnaver-ai\u002Fspeculative-forcing","speculative-forcing-video-denoising","2026-10-16T19:11:48Z","2026-10-16T19:11:52.976870Z","2026-10-16T19:11:52.976877Z",true,"agent",124,[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},"b950b487-2b1f-4ece-ad6e-d57cf94f1f84","稀疏注意力新突破：「上下文混合」让长视频生成成本降至近线性","moc-context-mixing-near-linear-video","2026-06-01T01:15:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"94894abf-62aa-41a9-8e3c-e999ff274d60","Sparse Forcing：稀疏注意力让视频生成质量速度双提升","meta-ucsb-sparse-forcing-pbsa-video-1-27x","2026-05-07T08:10:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"0b0af156-7b5c-47d8-bfc2-de8cc0b22ce2","混元开源Prism:稀疏注意力让2K音视频训练提速2.5倍","prism-2k-video-audio-sparse-attention","2026-10-13T23:13:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"24d56562-4886-4169-8aa9-773f5fa34e2f","GRACE 把视频潜空间压 8 倍:提速 11 倍,跑分反超","grace-latent-compression-video-diffusion","2026-10-08T15:10:59+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"40d7bf26-810e-4286-900f-f343fdb28ad5","ID-Forcing:免训练把5秒视野外推到240秒","id-forcing-long-video-kv-provenance","2026-10-07T23:09:14+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"]