[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-fasth3-4-step-distilled-minimax-h3":3,"topics-all":42,"news-related-02c8b500-ec11-44a6-8c58-6e880563dad8":60},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":28,"news_slug":35,"published_at":36,"created_at":37,"modified_at":38,"is_published":39,"publish_type":40,"image_url":14,"view_count":41},"02c8b500-ec11-44a6-8c58-6e880563dad8","FastH3 开源:4 步蒸馏版 MiniMax H3,B200 单卡最高提速 14 倍","UCSD FastVideo 团队联合 NVIDIA FastGen、Nuva Lab 开源 FastH3 Preview v1:用 DMD2 蒸馏把 MiniMax H3 的 49 次 DiT 调用压到 4 次,叠加 90% 稀疏注意力,B200 单卡最高提速 14 倍,8 卡 15 秒视频不足 13 秒生成。","视频生成的开源生态最近出现了一个新节奏:厂商放出基础权重,社区接力把推理成本打下来。MiniMax 今年夏天公开了 H3-Base 的模型权重,几周之后,UCSD 的 FastVideo 团队联合 NVIDIA FastGen 团队和 Nuva Lab 交出了第一份像样的答卷——FastH3 Preview v1,一组基于 MiniMax H3 后训练的开源权重,把文本生成音视频(T2VA)的速度抬到了新的水位线。\n\n## 49 次调用压到 4 次:DMD2 加 VSA 的组合拳\n\n先看问题有多大。按照官方博客的说明,Base H3 的音视频扩散 transformer 有 33B 参数,生成一个视频要完整调用它 49 次。FastH3 的思路是把这两个成本同时砍掉。\n\n第一刀是 DMD2(Distribution Matching Distillation)。学生模型用一个冻结的 Base H3 教师和一个学习到的 critic 来训练,两者 score 估计的差值提供训练信号;prompt-only 训练时用 backward simulation 让学生提前见识推理时会遇到的少步状态。效果是推理时调用次数从 49 次降到 4 次——五个调度点,恰好四次 DiT 前向。\n\n第二刀是 VSA(Video Sparse Attention)。可训练的稀疏注意力让学生只保留约 10% 的 video-to-video tiles(90% 稀疏度,64-token 块),文本和音频路径保持 dense;教师和 critic 也用全注意力,给稀疏学生一个完整注意力的目标。两刀叠加,同一个 35B 的 checkpoint,每次调用的计算密度也大幅下降。\n\n## 实测数字:B200 单卡最高 14.38 倍\n\n官方给出的延迟表很有信息量。1344×768、24 FPS 带音频的场景下,Base H3 Dense FA4 生成 15 秒视频在单张 B200 上要 678.7 秒;FastH3 推荐的 VSA\u002FData-Free checkpoint 只要 47.2 秒,加速 14.38 倍。5 秒视频是 132.5 秒对 16.2 秒,8.16 倍。4 张 B200 上,5 秒视频端到端 6.1 秒;8 张 B200 上,15 秒视频端到端 12.88 秒——不足 13 秒,已经低于视频本身的时长。这些数字的口径是预热后取三次计时中位数,端到端时间含编码、去噪、解码、音频、封装和落盘。\n\n训练成本方面,官方披露用了超过 1000 个 B200 训练小时。推荐 checkpoint 训练到 step 1300,同时提供全量权重和预提取的 rank-64 LoRA 两种形态。\n\n## data-free 训练:不用一条目标视频\n\n这批 checkpoint 有个值得注意的细节:推荐版本是 data-free 的,即训练只用 prompt、不用目标视频。这既绕开了视频数据的获取和版权问题,也解释了为什么这个蒸馏版本可以完全开放下载。官方同时发布了合成数据路线的消融版本(用 Base H3 自己生成的视频做训练源)和 dense 注意力参照版本,四个 checkpoint 组成完整的 Preview v1 家族,训练代码也承诺后续放出。\n\n限制也说得很直白:Preview v1 只支持 T2VA,首末帧到视频(FL2VA)和参考到视频(Ref2VA)都还没蒸馏;困难动作、细节和部分音频质量可能低于 Base H3;许可证沿用 MiniMax H3 Community License,商用前需要读一遍条款。\n\n## 为什么这件事比「又一个加速版」重要\n\n第一,它验证了开源权重的「二层创新」节奏。MiniMax 放出 H3-Base 权重后,社区可以检查它、后训练它、替换 kernel、在自己的硬件上跑它——这是闭源 API 模型做不到的。同一周 fal 发布的 H3 Max 走的是托管 API 路线,两条路线现在有了正面参照。\n\n第二,蒸馏加稀疏注意力这套组合拳,正在成为视频生成走向生产的标准路径。49 次调用压到 4 次,再叠加 90% 的注意力稀疏,意味着同样的硬件可以服务多得多的请求,或者反过来,同样的任务可以塞进小得多的机器。官方下一步计划还包括 8-step 质量版、FP8\u002FNVFP4 量化、面向 RTX、DGX Spark 和 Apple MLX 的本地优化,以及与 NVIDIA FastGen 合作的新蒸馏方法 PDD。\n\n对开发者的建议很直接:如果你在评估开源视频生成栈,推荐的 VSA\u002FData-Free checkpoint 值得跑一遍——权重硬件无关,4 张 B200 是官方的受控基准平台而非硬性要求,GPU 数量只需能整除 H3 的 56 个注意力头。\n\n当生成时间跌破视频本身的时长,交互式创作和实时工作流就从演示变成了可以排期的事——这才是这组开源权重真正的分量。\n\n原文:https:\u002F\u002Fhaoailab.com\u002Fblogs\u002Ffasth3-preview\u002F","https:\u002F\u002Fhaoailab.com\u002Fblogs\u002Ffasth3-preview\u002F","1e6f72ed-2aff-467f-9f66-e815d8426183",[11,15,18,22,25],{"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":21,"color":14},"f3be854a-50f0-411c-893d-16d0df6def02","h3-series","MiniMax H3 专题：持续追踪 H3 的发布、开源、蒸馏与部署全链路",{"id":23,"name":24,"slug":24,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":26,"name":27,"slug":27,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[29],{"id":30,"lang":31,"title":32,"summary":33,"content":34},"8aabe7c4-d409-4361-a7c0-5c3abc0a998a","en","FastH3 Open-Sources a 4-Step MiniMax H3: Up to 14x on One B200","FastVideo open-sources FastH3 Preview v1: DMD2 distillation cuts MiniMax H3's 49 DiT calls to 4, plus 90% sparse attention — up to 14x on one B200.","The open-source video generation stack just gained a second layer. Weeks after MiniMax released the H3-Base weights, the FastVideo team at UCSD, working with NVIDIA's FastGen team and Nuva Lab, shipped FastH3 Preview v1 — a family of open-weight checkpoints post-trained on MiniMax H3 that push text-to-audio-video (T2VA) generation into a new latency class.\n\n## From 49 transformer calls down to 4\n\nThe cost problem is stark. Base H3's audio-video diffusion transformer has 33B parameters, and generating one video invokes it 49 times. FastH3 attacks both the number of calls and the work inside each call.\n\nThe first cut is DMD2 (Distribution Matching Distillation). A frozen Base H3 teacher and a learned critic train the student; the difference between their score estimates supplies the training signal, and backward simulation exposes prompt-only students to the few-step states they will see at inference. The result: five scheduler points, exactly four DiT forwards at generation time.\n\nThe second cut is VSA (Video Sparse Attention). The trainable sparse-attention student keeps about 10% of eligible video-to-video tiles (90% sparsity, 64-token blocks) while text and audio stay dense; the teacher and critic also run dense attention, giving the sparse student a full-attention target.\n\n## The numbers: up to 14.38x on a single B200\n\nAt 1344×768, 24 FPS with audio, Base H3 with dense FA4 takes 678.7 seconds for a 15-second clip on one B200; the recommended VSA\u002FData-Free checkpoint takes 47.2 seconds — a 14.38x speedup. A 5-second clip: 132.5s vs 16.2s, 8.16x. On 4× B200 a 5-second clip lands at 6.1 seconds end-to-end; on 8× B200 a 15-second clip lands at 12.88 seconds — faster than the clip itself plays. Timings are the median of three requests after warmup, with model load and compilation excluded, and end-to-end includes encoding, denoising, decoding, audio, muxing, and file output.\n\nTraining consumed 1k+ B200 hours. The recommended checkpoint was trained to step 1300 and ships both as full weights and as a pre-extracted rank-64 LoRA.\n\n## Data-free training, stated limits\n\nThe recommended checkpoint is data-free: trained from prompts only, without target videos. That sidesteps video-data sourcing and licensing headaches entirely. The team also publishes synthetic-data ablations and a dense-attention reference — four checkpoints in the Preview v1 family, with training code promised soon.\n\nThe limits are stated plainly: T2VA only (FL2VA and Ref2VA are not distilled yet), difficult motion, fine detail, and some audio may fall below Base H3 quality, and everything inherits the MiniMax H3 Community License.\n\n## Why this matters beyond \"another speedup\"\n\nIt validates the two-layer innovation model that open weights enable: MiniMax shipped the base, and the community now inspects it, post-trains it, swaps kernels, and runs it on its own hardware. fal's H3 Max, released the same week, took the hosted-API route — the two paths now have a direct comparison.\n\nDistillation plus sparse attention is becoming the standard route to production video generation. 49 calls down to 4, plus 90% attention sparsity, means the same hardware serves far more requests — or the same task fits into much smaller machines. Next on the roadmap: an 8-step quality variant, FP8\u002FNVFP4 quantization, local optimizations for RTX, DGX Spark, and Apple MLX, and a new Parallel Decoding Distillation (PDD) collaboration with NVIDIA.\n\nIf you are evaluating open video stacks, run the recommended VSA\u002FData-Free checkpoint: the weights are hardware-independent, the 4× B200 setup is a controlled benchmark rather than a requirement, and the GPU count just needs to divide H3's 56 attention heads.\n\nWhen generation time drops below the video's own duration, interactive and real-time workflows stop being demos and start being schedulable work. That is the real weight of this release.\n\nSource: https:\u002F\u002Fhaoailab.com\u002Fblogs\u002Ffasth3-preview\u002F","fasth3-4-step-distilled-minimax-h3","2026-08-30T21:30:00Z","2026-08-30T21:12:36.694350Z","2026-08-30T21:12:36.694366Z",true,"agent",395,[43,52],{"slug":44,"tag_slug":44,"title_zh":45,"title_en":46,"intro_zh":47,"intro_en":48,"id":49,"is_active":39,"created_at":50,"modified_at":51},"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":20,"tag_slug":20,"title_zh":53,"title_en":54,"intro_zh":55,"intro_en":56,"id":57,"is_active":39,"created_at":58,"modified_at":59},"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":61},[62,67,72,77,82,87],{"id":63,"title":64,"news_slug":65,"published_at":66},"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":68,"title":69,"news_slug":70,"published_at":71},"18d2aa73-7244-4b10-b611-46475e17327e","ForgeWM开源:一步去噪72FPS的可玩世界模型,8张卡复现全流程","forgewm-few-step-playable-world-model","2026-08-24T21:10:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"0599b775-ac17-49d2-aebd-a16f531c7168","腾讯混元 MeanFlowNFT：把 RL 接进「平均速度生成器」，Wan 2.1 4 步反超 50 步 LongCat-Video RL","tencent-hunyuan-meanflownft","2026-07-16T12:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"42cfc778-8f1b-4bf2-a0ae-4343a066f48d","RhymeFlow：清华提出异步去噪流调度，DiT视频生成训练免费加速1.53倍","rhymeflow-tsinghua-async-denoising-1-53x","2026-06-07T22:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"bf8755fc-cd4f-4bd9-9617-e70f56ddc4ac","LTX-2.3：开源视频生成正式进入 4K + 原生音频时代","ltx-2-3-lightricks-4k-native-audio","2026-06-02T01:00:00+00:00",{"id":88,"title":89,"news_slug":90,"published_at":91},"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"]