[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-id-forcing-long-video-kv-provenance":3,"topics-all":38,"news-related-40d7bf26-810e-4286-900f-f343fdb28ad5":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},"40d7bf26-810e-4286-900f-f343fdb28ad5","ID-Forcing:免训练把5秒视野外推到240秒","首尔大学等四校团队提出 ID-Forcing:自回归视频扩散的长视频漂移,病根在 KV 缓存的\"溯源\"失配——训练视野之外的缓存条目本身已是分布外数据,事后挑读救不回来。self-caching 从写入端杜绝越界,配合精确滚动窗口,免微调把 5 秒视野外推到 240 秒,两项漂移指标大幅超越既有基线,代码开源。","自回归(AR)视频扩散模型有个老毛病:视频一长,颜色和纹理慢慢跑偏,运动幅度逐渐衰弱,业内称之为漂移(drifting)。主流修法是 KV conditioning——推理时挑选或改写缓存里的 key-value 条目,控制模型\"读\"什么。首尔大学、高丽大学、成均馆大学与佐治亚理工四校团队给出了不同诊断:光管\"读\"不够,病根在\"写\"。论文已发布于 arXiv(https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.03120),代码同步开源。\n\n## KV 溯源:条目写入时就已越界\n\n团队的观察很直接:KV conditioning 隐含假设缓存条目仍在训练分布内,但该假设在训练视野之外失效。一个 KV 条目依赖它被缓存时注意到了哪些条目——即它的溯源(provenance)。模型只在短视野上训练,推理 rollout 时每个新 chunk 都是在训练从未执行过的溯源状态下写入缓存,条目本身就成了分布外(OOD)数据。团队将其命名为 KV-provenance problem,推论相当残酷:溯源在写入那一刻就已固定,事后任何 conditioning 策略都无法挽回。\n\n## 两级规则:写入端自缓存,读取端精确窗口\n\nID-Forcing 是纯测试时框架,零微调,规则只有两条:\n\n- **写入端 self-caching**:窗口内最早的条目只注意自己写入,其余条目在其之上自回归写入。每条 KV 的写入溯源都与训练执行过的状态一致,从源头杜绝 OOD。\n- **读取端精确滚动窗口**:永久保留首个 chunk κ0(训练每一步都以它为条件),其余只留最近条目并做时间距离再旋转,窗口严格满足 L ≤ N−1,滚动窗口精确落在分布内。\n\n## 效果:免训练外推 48 倍时长\n\n项目页演示中,单一提示词生成的视频被外推到 30\u002F60\u002F120\u002F240 秒,而基座训练视野约 5 秒,全程无微调。仓库默认脚本生成 2 分钟视频,即 160 个 chunk、1917 帧(16 fps)。评测覆盖两个基座:Self-Forcing 从第 7 个 chunk(跨过其 7-chunk 训练视野的第一个 chunk)起接管;LongLive 从第 1 个 chunk 起接管,它本身就用分钟级视频训练。结果显示方法在标准视频生成基准上保持竞争力,缓解漂移则大幅超越此前工作——由两项漂移指标与用户研究共同验证:颜色漂移度量首尾帧 180-bin HSV 色调直方图差异,运动漂移用 VBench Dynamic Degree(首尾各 5 秒,RAFT 光流、8 fps 采样)。对比基线包括 Self-Forcing、Deep Forcing、MemRoPE 与 Infinity-RoPE。\n\n## 开源与复现门槛\n\nGitHub 仓库结构干净:run_self_forcing.sh 与 run_longlive.sh 两条一键脚本,权重表写明 Self-Forcing 用 EMA 权重、LongLive 用 base+LoRA,文本编码器与 VAE 取自 Wan2.1-T2V-1.3B;评测提示词为 128 条 MovieGen prompt;代码沿用两个基座原仓库的 Apache License 2.0。断点续跑开箱即用——已存在的视频自动跳过。\n\n## 所以呢\n\n这件事的价值不在\"又把视频拉长了\",而在成本结构:分钟级视野意味着分钟级数据与算力,而 ID-Forcing 证明只要推理期把 KV 的读写都对齐训练分布,5 秒视野的现成模型就能稳定外推。对算力有限的团队,这是\"用推理期工程换训练期投入\"的又一个样本;它也提醒整个社区:KV cache 不只是省显存的工程件,它本身就是分布问题的载体——文本推理侧的缓存研究热,与视频生成侧的这条线索,正在说同一件事。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.03120","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},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"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},"65715e39-a98f-4882-89ff-fc1bcf16c77c","en","ID-Forcing Stretches 5s Video Models to 240s, Training-Free","ID-Forcing traces long-video drift to OOD KV cache; self-caching plus exact rolling windows stretch 5s-horizon models to 240s training-free.","Autoregressive (AR) video diffusion models share an old ailment: as clips get longer, colors and textures slowly shift and motion decays — the field calls it drifting. The dominant fix is KV conditioning, which selects or rewrites cached key-value entries at inference to control what the model reads. A four-school team from Seoul National University, Korea University, Sungkyunkwan University and Georgia Tech offers a different diagnosis: managing what is read is not enough, because the failure starts at how entries are written. The paper is on arXiv (https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.03120), with code released alongside.\n\n## KV provenance: entries go out-of-distribution at write time\n\nThe observation is blunt: KV conditioning implicitly assumes cached entries stay in-distribution, and the assumption breaks beyond the training horizon. A KV entry depends on the entries it attended to when it was cached — its provenance. Trained only on short horizons, the model writes every new chunk during rollout under a provenance never executed in training, so the entries themselves become out-of-distribution (OOD). The team names this the KV-provenance problem, with a harsh corollary: provenance is fixed at caching time, and no conditioning policy can recover from it.\n\n## Two rules: self-caching on write, an exact window on read\n\nID-Forcing is a pure test-time framework — no fine-tuning — with one rule per KV operation:\n\n- **Write side, self-caching**: the earliest entries in the window are cached attending only to themselves, the rest autoregressively on top. Every entry is written under a provenance training actually executed, preventing OOD at the source.\n- **Read side, exact rolling window**: the first chunk k0 — which every training step conditions on — is permanently kept, alongside the most recent entries, re-rotated so temporal distances stay in the trained range, with window length strictly L \u003C= N-1.\n\n## Results: a 48x training-free stretch\n\nOn the project page, videos from a single prompt are extrapolated to 30\u002F60\u002F120\u002F240 seconds from a roughly 5-second training horizon, with no fine-tuning. The repo's default scripts generate 2-minute videos — 160 chunks, 1917 frames at 16 fps. Evaluation spans two backbones: ID-Forcing takes over Self-Forcing at chunk 7 (the first chunk past its 7-chunk training horizon) and LongLive at chunk 1 (itself trained on minute-long videos). The method stays competitive on standard video-generation benchmarks while substantially outperforming prior work on drift, validated by two self-built drift metrics and a user study: color drift compares 180-bin HSV hue histograms of the first and last frames; motion drift uses VBench Dynamic Degree over the first and last 5 seconds (RAFT optical flow, 8 fps subsampling). Baselines include Self-Forcing, Deep Forcing, MemRoPE and Infinity-RoPE.\n\n## Open source and reproduction\n\nThe GitHub repo is clean: two one-shot scripts drive Self-Forcing and LongLive; the checkpoint table specifies EMA weights for Self-Forcing, base+LoRA for LongLive, with the text encoder and VAE from Wan2.1-T2V-1.3B; evaluation uses 128 MovieGen prompts; code keeps the Apache License 2.0 of both base repos. Interrupted runs resume out of the box — existing videos are skipped.\n\n## So what\n\nThe value is not \"longer videos again\" but the cost structure: minute-scale horizons demand minute-scale data and compute, while ID-Forcing shows that aligning KV reads and writes with the training distribution at test time lets an off-the-shelf 5-second model extrapolate stably. For compute-limited teams it is another case of trading inference-time engineering for training-time spend — and a reminder that the KV cache is not just a memory-saving artifact but the very carrier of a distribution problem. The hot cache research on the text side and this thread on the video side are telling the same story.","id-forcing-long-video-kv-provenance","2026-10-07T23:09:14Z","2026-10-07T23:09:19.808254Z","2026-10-07T23:09:19.808264Z",true,"agent",181,[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},"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":65,"title":66,"news_slug":67,"published_at":68},"ab13d62e-1d32-4af6-9d11-ebdb601ff274","Kandinsky 6.0 开源:29B 音视频一体生成","kandinsky-6-0-video-audio-generation-open-source","2026-10-06T19:15:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"af21d26d-d9cd-4526-ac4a-66366a45848c","AV-GRPO:8张A800给22B音视频模型做RL后训练","av-grpo-audio-video-diffusion-rl","2026-09-27T17:08:13+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"9210c86b-3e1d-434b-bbfd-78e62c698aed","WorldCrafter 开源:给视频世界模型装上可查询的 3D 记忆,转一圈回来还是那个房间","worldcrafter-video-world-model-3d-memory","2026-09-22T19:08:48+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"caed836e-2168-418f-b5c1-bde3ce962e66","Mask Forcing 往蒸馏 rollout 里掺干净 token:修视频生成的模式坍缩,指令遵循最高涨 6.5 分","mask-forcing-video-diffusion-distillation","2026-09-09T23:08:37+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"fb2da954-b12e-4bde-9146-b61dd240df92","SolarWM 开源:143 万条视频喂出的世界模型,5 秒训练片段撑起小时级交互","solarwm-open-data-video-world-models","2026-09-03T15:08:13+00:00"]