[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-vc-attention-low-bit-video-attention":3,"topics-all":38,"news-related-ff6f65e1-28b2-4a48-b317-7870072ecfa9":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},"ff6f65e1-28b2-4a48-b317-7870072ecfa9","VC-Attention低比特注意力:视频生成提速1.59倍","Nunchux AI提出免训练低比特注意力框架VC-Attention,五卡四视频模型实测注意力内核提速1.46-1.59倍,保真度优于同类。","视频生成模型的效果一年一个台阶,但推理账单也在同步变厚。Nunchux AI 在官方博客里给了一个直观的数字:用 BF16 FlashAttention-4 在一块 B200 上跑 MiniMax-H3,每个去噪步约有三分之二的时间花在注意力上;视频 token 数越长,这笔开销按平方增长。注意力,就是视频扩散模型部署成本的咽喉。\n\n## 低比特注意力的两堵墙\n\n低比特化是显而易见的省钱路径,但会撞上两堵墙。第一堵是精度:量化以硬件块为单位设 scale,scale 由块内最大的条目决定,离群值一出现,普通条目就被挤进很窄的可表示区间。此前的工作已经能平滑 query 和 key,但 value 的离群值不遵循固定的通道或时空结构,成了输出误差的主要来源。第二堵是速度:低比特 Tensor Core 只加速注意力里的两次矩阵乘,夹在中间的 softmax 仍要跑 FP32 高精度指数运算,反而成了内核流水线上最长的一段。\n\n## V-Smooth 与 ExpCast-FP8\n\nVC-Attention 的解法是两面各拆一堵墙。V-Smooth 用轻量在线 k-means 把 value token 重新分组,让同一个硬件块里的 token 彼此相近;再减掉块均值、只量化残差,均值则从在线 softmax 本来就要维护的行和里恢复。论文称,仅这一步就在四个模型上比 SageAttention2 多拿回 1.1 到 2.8 dB 的 PSNR。ExpCast-FP8 则把 log 域分数用一次融合乘加直接映射成 E4M3 概率编码,把 FP32 指数运算和格式转换整个抹掉。\n\n## 实测数字\n\n论文在 B200、B300、H200、RTX PRO 6000、RTX 5090 五种 GPU 上实现并评测,覆盖 Wan2.2、LongCat-Video、HunyuanVideo-1.5、MiniMax-H3 四个视频模型。注意力内核相对 BF16 FlashAttention-4:数据中心卡提速 1.46-1.59 倍,工作站卡 2.3-3.6 倍;端到端出片分别提速 1.13-1.19 倍与 1.36-1.70 倍。MiniMax-H3 生成 243 帧 1344×768 的负载上,B200 内核提速 1.59 倍、B300 为 1.51 倍;100 个 prompt 的保真度均值 20.2 dB,略优于 SageAttention2 的 19.9 dB。\n\n## 免训练才是要害\n\n论文出自 Nunchux AI,作者名单包括 Song Han、Yujun Lin、Lvmin Zhang 等。对工程侧来说,最大的卖点不是某个单项倍数,而是 training-free:不重训、不动权重,现有视频模型直接换注意力内核就能吃到加速,还能与稀疏注意力、少步蒸馏、多卡执行叠加。当视频生成从演示走向产能,每一分推理成本都会被放大千万次——低比特这条路绕不开 softmax 这道坎,VC-Attention 把坎铲平了一块。\n\n所以呢?模型军备赛看参数,成本战看内核。谁的注意力先跑进 8 比特时代,谁的视频 API 报价单就更有底气。[论文原文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.15810)与[Nunchux 官方博客](https:\u002F\u002Fwww.nunchux.ai\u002Fblog\u002Fattention-is-the-video-bottleneck)均已公开,数字值得自己核一遍。\n","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.15810","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",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},"b49648f9-963e-4082-8684-3d085b7358fe","quantization",{"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},"c7040bc8-7eb1-4c90-9503-3574d785c812","en","VC-Attention: Training-Free Low-Bit Video Attention","Nunchux AI's VC-Attention removes the FP32 softmax and quant error in low-bit video attention, speeding kernels 1.59x without retraining.","Video generation models improve year over year, and so does the inference bill. Nunchux AI's blog gives a concrete number: running MiniMax-H3 with BF16 FlashAttention-4 on a single B200, about two thirds of every denoising step is spent on attention; the longer the clip, the more that cost grows with the square of the token count. Attention is the choke point of video diffusion deployment cost.\n\n## Two Walls of Low-Bit Attention\n\nLow-bit quantization is the obvious path to savings, but it hits two walls. The first is accuracy: a hardware block's quantization scale is set by its largest entries, so when outliers appear, typical entries get squeezed into a narrow representable range. Prior work already smooths queries and keys, but value outliers follow no fixed channel or spatiotemporal structure and remain the dominant source of output error. The second is speed: low-bit Tensor Cores accelerate only the two matrix multiplications; the softmax between them still runs in FP32, and that high-precision exponential becomes the longest pipeline stage of the kernel.\n\n## V-Smooth and ExpCast-FP8\n\nVC-Attention tears down both walls. V-Smooth uses lightweight online k-means to regroup value tokens so that tokens in the same hardware block are similar; it then subtracts the block mean and quantizes only the residual, restoring the mean from the row sum that online softmax already maintains. Across four models, this step alone adds 1.1 to 2.8 dB of PSNR over SageAttention2. ExpCast-FP8 maps log-domain scores directly to E4M3 probability codes with one fused multiply-add, eliminating the FP32 exponential and the format conversion entirely.\n\n## Benchmark Numbers\n\nThe paper is implemented and evaluated on five GPUs — B200, B300, H200, RTX PRO 6000, and RTX 5090 — across four video models: Wan2.2, LongCat-Video, HunyuanVideo-1.5, and MiniMax-H3. Relative to BF16 FlashAttention-4, the attention kernel speeds up 1.46-1.59x on datacenter Blackwell and Hopper cards and 2.3-3.6x on workstation cards; end-to-end clip generation is 1.13-1.19x and 1.36-1.70x faster respectively. On the MiniMax-H3 workload generating 243 frames at 1344×768, the kernel is 1.59x faster on B200 and 1.51x on B300; fidelity across 100 prompts averages 20.2 dB PSNR, slightly ahead of SageAttention2's 19.9 dB.\n\n## Training-Free Is the Point\n\nThe paper comes from Nunchux AI, with authors including Song Han, Yujun Lin, and Lvmin Zhang. For engineers, the biggest selling point is not any single multiplier but the fact that it is training-free: no retraining, no weight changes — existing video models can swap in the attention kernel and take the speedup, and it composes with sparse attention, few-step distillation, and multi-GPU execution. As video generation moves from demo to production, every bit of inference cost gets multiplied millions of times over. The low-bit road cannot bypass softmax, and VC-Attention flattens a piece of that barrier.\n\nSo what? Model races are about parameters; cost wars are about kernels. Whoever moves attention into the 8-bit era first gets the more confident pricing sheet for their video API. The [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.15810) and the [Nunchux blog post](https:\u002F\u002Fwww.nunchux.ai\u002Fblog\u002Fattention-is-the-video-bottleneck) are both public — worth verifying the numbers yourself.\n","vc-attention-low-bit-video-attention","2026-09-17T13:30:00Z","2026-09-17T13:16:39.746985Z","2026-09-17T13:16:39.746994Z",true,"agent",18,[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},"178aa5e5-2a4f-4a87-a97c-0da16295d96f","EMNLP 2026 OCGQuant:用通道配对治 NVFP4 陪葬误差,Qwen3-1.7B 接近 FP16","ocgquant-nvfp4-outlier-companion-grouping","2026-09-10T09:15:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"4147f71b-eaa7-4d39-91cf-c2c572105e7f","FlashPrefill V2:128K 长文本 prefill 提速 47 倍,块稀疏注意力走进生产框架","flashprefill-v2-block-sparse-prefill","2026-08-21T19:10:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"c32d3160-4e07-4128-890f-4e135aac2cce","CompactifAI 把 Llama 3.3 70B 砍到一半:Multiverse 在 Intel Xeon 6 上跑出 1.9 倍吞吐","compactifai-llama-3-3-70b-intel-xeon","2026-07-26T04:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"cf7f8e7e-4efc-451e-9595-706b0be911ba","PolyQ:把\"3-bit LLM 跑在 CPU\"做成一件可预测的事","polyq-3bit-llm-cpu","2026-07-17T10:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"007a83c4-faed-459a-ab44-17b915e05fb5","TriRoute 把 MoE + MoD + KV 量化做成一个控制器：三条条件计算路径第一次协同","triroute-moe-mod-kv-quantization","2026-07-09T18:02:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"6452bb36-79c8-471b-aa4e-fad99bca9b04","SharQ 用「稀疏-稠密双轨」把 FP4 推理提速 2.4 倍:训练免费还跨平台","sharq-sparse-dense-fp4","2026-07-01T00:00:00+00:00"]