全局键值缓存每token仅八百九十字节,仓库修复升至74.2%

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

DeepSeek-AI, :, Anyi Xu, B. Li, Bangcai Lin, Bing Xue, BingCheng Xian, Bingzheng Xu, Bochao Wu, Bowei Zhang, Boyi Deng, C. C. Yu, Chao Jin, Chaofan Lin, Chen Dong, Chenbing Wang, Chenfan Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chengyuan Zhang, Chenhao Xu, Chenqi Zhao, Chenze Shao, Chuhao Wang, Chuqi Zhang, Damai Dai, Dejian Yang, Deli Chen, Di Huang, Di Wu, Donghao Li, Erhang Li, Eric Fu, F. Zhou, Fangwei Zhou, Fangyun Lin, Fangzhou Yuan, Feiyu Xia, Fucong Dai, Guangbo Hao, Guanglin Li, Guanting Chen, Guoai Cao, Guofan Fan, Guolai Meng, Guowei Li, Haichuan Zhang, Haiyang Ma, Haiyang Shen, Han Li, Han Yu, Han Zhang, Hangyuan Deng, Hanwei Xu, Hanxiang Xu, Hanxun Zhong, Hao Guo, Hao Jiang, Hao Li, Hao Qin, Haodong Wen, Haofen Liang, Haofeng Huang, Haohua Liu, Haoling Zhang, Haoming Luo, Haoran Yang, Haotian Xu, Haotian Yuan, Haoting Huang, Haowen Luo, Haoyang Cai, Haoyu Chen, Haozhe Ji, Hengran Zhang, Hengrui Wang, Hengxu Wu, Honghui Ding, Hongxuan Tang, Huadong Wang, Huanqi Cao, Huazuo Gao, Hui Qu, Hui Zeng, J. Yang, J. H. Jin, J. H. Zhang, J. X. Zou, Jia Yu, Jiahui Zhou, Jiajun Chen, Jialiang Huang, Jialin Zhao, Jiamin Tang, Jian Zhou, Jianan Tong, Jianwen Li, Jiaqi Zhu, Jiarui Wang, Jiasheng Ye, Jiashi Li, Jiaxin Xu, Jiaying Ding, Jibai Lu, Jiewen Hu, Jin Yan, Jincheng Zhai, Jingchang Chen, Jingcheng Hu, Jingli Zhou, Jingsheng Xu, Jingting Xiang, Jingyan Yun, Jingyang Yuan, Jingyuan Cheng, Jinhua Zhu, Jinpeng Wang, Jinyi Chen, Jinyi Hu, Jiping Yu, Jueliang Guo, Junbo Pei, Junbo Sun, Junguang Jiang, Junjie Qiu, Junkang Zhou, Junqi Liu, Junren Li, Junxian Li, Junxiao Song, Junyi Guo, Kai Dong, Kaifeng Chen, Kaige Gao, Kang Guan, Kangdong Yuan, Ke Hong, Ke Xu, Kefan Zhao, Kexin Ji, Kexin Zhang, Kexing Zhou, Kuai Yu, Lan Zhang, Lean Wang, Lecong Zhang, Lei Wang, Letian Gao, Liang Zhao, Liansheng Xu, Lihua Guo, Lingxiao Luo, Lingyue Fu, Litao Deng, Litong Wang, Liyue Zhang, Longhao Chen, Lu Chen, Luotian Huang, Luyao Ma, Luyao Wang, M. S. Di, Max Mei, Menghao Ye, Miao Cui, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Mingjing Zhang, Mingqi Wei, Mingshu Chen, Mingxing Liu, Mingxu Zhou, Mingyu Xu, Mingyu Yang, Mingze Wang, Muyang Chen, Ni Shentu, Ning Wang, Niufang Ning, Panpan Huang, Peixin Cong, Peiyi Wang, Peiyuan Xin, Pengfei Ren, Pengfei Yan, Pengle Zhang, Qi Kang, Qi Tang, Qiancheng Wang, Qiang Li, Qihao Zhu, Qingyang Li, Qinyu Chen, Qiushi Du, Qizhou Guo, Rongxian Xu, Rui Ding, Rui Hu, Rui Tian, Rui Yu, Ruidong Zhu, Ruifan Xu, Ruihan Yang, Ruihang Xia, Ruijie Lu, Ruilin Geng, Ruipeng Hong, Ruiqi Ge, Ruisong Zhang, Ruize Sun, Ruizhe Pan, Runji Wang, Runqian Chen, Runxin Xu, Ruohong Tian, Ruomeng Shen, Ruoyu Zhang, Ryan X., S. H. Liu, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shaofei Cai, Shaoheng Nie, Shaoyuan Chen, Shengding Hu, Shengkai Lin, Shengwen Ran, Shengyu Liu, Shengyuan Jia, Shi Bai, Shi Feng, Shicheng Xu, Shichun Liu, Shiqiang Hu, Shirong Ma, Shiyu Wang, Shiyuan Feng, Shufan Gong, Shuhan Lin, Shuiping Yu, Shunfeng Zhou, Shuo Yang, Shuomeng Wang, Shuting Guo, Shuting Pan, Shuying Yu, Sinuo Cao, Siyi Lin, Sizhe Chen, Songyang Chen, Songyang Zhou, Tao Ni, Tao Yun, Tian Jin, Tian Pei, Tian Ye, Tianle Lin, Tianran Ji, Tianyi Cui, Tianyuan Yue, Tingting Yu, Tongrui Xiong, Wangding Zeng, Wei Liu, Wei Zhang, Weibin Xu, Weihao Zeng, Weilin Zhao, Wen Liu, Wenfeng Liang, Wenjie Pang, Wenjing Luo, Wenjing Yao, Wenjun Gao, Wenkai Shao, Wenkai Yang, Wenli Zhang, Wenlu Wang, Wenlve Huang, Wenqian Yan, Wentao Zhang, Xi Gao, Xiang He, Xiang Li, Xiangli Li, Xiangwen Wang, Xiangying Zhang, Xiankui Wei, Xiao Bi, Xiaodong Liu, Xiaohan Wang, Xiaojian Qu, Xiaokang Chen, Xiaokang Zhang, Xiaotao Nie, Xiaoyao Zou, Xiaoyuan Li, Xicheng Guo, Xieting Chu, Xin Cheng, Xin Liu, Xin Xie, Xinbo Xu, Xingchao Liu, Xingchen Liu, Xingkai Yu, Xingyou Li, Xintong Yao, Xinyang Chen, Xinyong Jiang, Xinyu Yang, Xinyu Yang, Xu Chen, Xuanyu Wang, Xubei Zhong, Xuecheng Su, Xuejie Liu, Xuheng Lin, Xujie Fan, Xuncheng Zhao, Xuwei Fu, Y. C. Yan, Y. H. Jiang, Y. T. Wu, Y. W. M., Y. Z. Wang, Yafei Gao, Yang Yang, Yang Zhang, Yanru Ma, Yanwen Huang, Yao Li, Yao Li, Yao Meng, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yaoyang Ye, Yehang Yin, Yexinrui Wu, Yi Qian, Yi Tao, Yi Yu, Yichao Zhang, Yichen Jiang, Yicheng Wang, Yifan Ding, Yifan Shi, Yifeng Peng, Yifeng Zhai, Yijia Wu, Yiliang Xiong, Yilun Wang, Ying He, Ying Zhou, Yingjia Luo, Yinmin Zhong, Yiping Wang, Yisong Wang, Yixiang Zhang, Yixiao Chen, Yixuan Tan, Yixuan Wei, Yiyang Ma, Yiyao Yang, Yiyuan Liu, Yizai Cai, Yizhen Wei, Yizhi Wang, Yonglun Yang, Yongqi Zhuo, Yongqiang Guo, Yongtong Wu, Yu Wu, Yu Zhang, Yuan Bian, Yuan Cheng, Yuan Ou, Yuan Sun, Yuanfan Xu, Yuanhang Sun, Yuanhao Li, Yuchen Liu, Yuchen Yao, Yudong Han, Yuduan Wang, Yuhan Wu, Yuhao Meng, Yuheng Zou, YuKun Li, Yunchuan Wang, Yunfan Xiao, Yunfan Xiong, Yupeng Chen, Yuqian Cao, Yuqian Wang, Yuqing Chen, Yushun Zhang, Yutong Lin, Yuwei Xiao, Yuxian Gu, Yuxiang Chen, Yuxiang Huang, Yuxiang Luo, Yuxiang You, Yuxin Chen, Yuxin Xiang, Yuxuan Liu, Yuxuan Zhou, Yuyang Zhou, Yuzhe Guo, Yuzhen Huang, Yuzhuo Bai, Z. Y. Z., Zanlin Ni, Zehao Wang, Zehua Zhao, Zehui Ren, Zejun Zhao, Zhangli Sha, Zhanying Wang, Zhaochen Zhang, Zhaoshuai Du, Zhe Fu, Zhean Xu, Zhenda Xie, Zheng Liu, Zhengyan Zhang, Zhenhua Dong, Zhewen Hao, Zhibang Wang, Zhibin Gou, Zhicheng Ma, Zhihao Li, Zhihong Shao, Zhihuan Huang, Zhijie Li, Zhirui Lu, Zhixian Huang, Zhixuan Chen, Zhixuan Chen, Zhixuan Pan, Zhiyu Wu, Zhizhou Ren, Zhu He, Zhuoshu Li, Zhuping Zhang, Zian Xu, Zihao Wang, Zihui Gu, Zijia Zhu, Zili Zhang, Zilin Li, Zilong Hou, Zilong Lyu, Ziqiao Wang, Ziwei Xie, Ziya Zhang, Ziyi Gao, Zizheng Pan, Zonglin Li, Zongqing Yao, Zui Chen, Zuofan Wu, Chenchen Ling, Chengyu Hou, Chong Chen, D. Li, Di Qi, Dongjie Ji, Fang Wei, Fanyi Xia, Fei Xie, Feiyi Tan, Hailong Guo, Haiyan Zhai, Hui Zhou, Huihui Tan, Huijie Li, Jia Luo, Jia Song, Jialu Cai, Jian Liang, Jiangting Zhou, Jiaqi Gao, Jiayi Shao, Jie Chen, Jieyu Yang, Jin Chen, Jingde Zhang, Jingzi Zhou, Jinqian Wang, Jinyang Liu, JinZhao Sun, Junhua Ling, Junmin Zheng, Kaicheng Yang, Ke Xu, Le Su, Leyi Xia, Liangfeng Ding, Lin Zhuo, Linwang Ma, Linyan Zhu, Liyu Cai, Luqi Yao, M. K. Zhang, Meng Li, Miao Lin, Miaojun Wang, Min Zhang, Mingming Li, Mingming Wang, Mingze Yin, Minmin Han, Nan Cao, Ning Wang, Ningxin Ma, Panpan Wang, Peihan Lin, Peng Sun, Peng Zhang, Qian Ying, Qiang Xiang, Qiao Wang, Qingmiao Mao, Qiwei Jiang, Rongli Jin, Ruyi Chen, Sha Tao, Shangmian Sun, Shaoqing Wu, Shichao Zou, Si Lei, Tianyang Zhang, Tianyu Sun, Tingting Yin, W. L. Xiao, Wei An, Wei Li, Wei Wang, Weiwei Lin, Wenqing Hou, X. Lin, Xiangfei Meng, Xianzhu Huang, Xiao Peng, Xiaoqian Li, Xiaoting Zhang, Xiaowen Sun, Xiaoxiang Wang, Xiaoyu Ye, Xinrou Zhang, Xinyu Zhang, Xue Cao, Xueyin Chen, Yanan Zhou, Yanhong Xu, Yao Xia, Yao Xu, Yi Shao, Yihong Zhang, Yiling Ma, Ying Tang, Yining Lou, Yiru Chen, Yishi Piao, Yixuan Chen, Yong Xiong, Yuchen Xuan, Yuehan Yang, Yuer Xu, Yukun Zha, Yunxian Ma, Yuping Lin, Yuting Yan, Yutong Xie, Yuwen Sheng, Yuxuan Zhu, Zekai Zhang, Zhe Ju, Zhenzhen Lin, Zheren Gao, Zheyang Sun, Zhigang Yan, Zhongyu Wu, Zi Wang, Zihua Qu, Ziling Yan, Ziyi Wan

cs.CL

2026-09-17

552B的MoE用编码器解码器切半、跨层复用稀疏注意力和FP4,把常驻显存的全局缓存压到每token 890字节,约是V4-Flash的四分之一;DeepSWE升到74.2%,命令行基准2.1到90.6%。

这篇在解决什么

长程 agent 把负载推向「输入很重」:prefill 贵,KV 占满 HBM,持久化再挤 SSD 和搬运带宽。DeepSeek-V4 已经用全局稀疏注意力加每层滑动窗(SWA)把计算压下去,超长序列上真正卡的是全局 KV 的存和搬。

V4.1-Flash 把压缩当成一等目标。骨干 552B、Engram 再加 196B,上下文到 100 万 token,45T 多模态语料预训练。Decode 每 token 激活 16B,prefill 只激活 8B。相对 V4-Flash,同长度下运行时全局 KV 大约 1/4,持久 KV 大约 1/8,分数还更高。

方法

因果编码器–解码器(CED)把 40 层对半切。解码器的全局 KV 从编码器最后一层隐状态投影,超长 prompt 只需跑前半层拿全局缓存,SWA 仍逐层算。SWA Bounded Replay 重放最近一个窗口的 token 来近似重建 SWA 状态,避免把 SWA KV 持久化到 SSD。

Compressed Sparse Attention 2(CSA2)沿条目大小、序列压缩、跨层复用三条轴同时动手。层有三种静态模式:Full 自己产全局 KV 和 Top-K;Reindex 复用 KV、用本层 indexer Q 重打分;Reuse 连 Top-K 一起复用。解码器里分层 indexer 让后续层只在更早层选出的候选池里打分,候选池大小固定后,深层索引成本不再随上下文线性涨。主 KV 用 FP4(E2M1,每 16 通道一个 E4M3 标度)做量化感知训练;SWA KV 仍 FP8。

另外还有 Single-Pass mHC(残差混合一次读完)、Engram 条件记忆、DSpark 推测解码。后训练按 SFT→RL→on-policy distillation 走,算法不标新,把力气放在可验证任务合成和交互环境扩容。

结果

全局 KV 打到每 token 890 字节。Decode FLOPs 随上下文几乎走平:4K 拉到 1M(256 倍)只增加约 1/4。

Agent 侧,相对 V4-Flash 的跳跃大于参数叙事。DeepSWE v1.1 从 54.4% 到 74.2%,略过 Opus-5 的 74.0% 和 GPT-5.6 Sol 的 73.0%。Terminal-Bench 2.1 从 82.7% 到 90.6%(Opus-5 89.1%)。Automation-Bench 54.8%、Agents’ Last Exam 31.8%,表内最高。Codeforces 评分 3471,高于 V4-Pro 的 3348。GPQA Diamond 90.9%。HLE 36.8(文本子集 39.1)仍低于若干更大模型。Terminal-Bench 4.0 为 31.2%,对 Opus-5 的 51.8% 仍有洞。

推理力度 25 拉到 100,八个推理基准均 Pass@1 从 67.1% 到 76.3%,DeepSWE 66.0%→74.2%,TB2.1 82.4%→90.6%,输出 token 大约 2.5 倍。60–80 档已经收回大部分准确率。多 agent 在 ProgramBench 上 8 小时 Almost@1 到 30.04%,单 agent 20.39%。

基座相对 V4-Pro-Base:约 1/3 总参数、1/4 激活,内部评测称世界知识与推理编码接近,held-out 语料 BPB 全面更低。论文也写:日常任务体验接近前沿闭源,最难推理和边角仍有差距。

基准V4.1-FlashV4-FlashOpus-5
DeepSWE v1.174.2%54.4%74.0%
Terminal-Bench 2.190.6%82.7%89.1%
Terminal-Bench 4.031.2%7.0%51.8%
Automation-Bench54.8%37.7%50.3%
GPQA Diamond90.9%89.9%93.4%

为什么重要

这是部署论文多于「再堆一个更大 MoE」。Agent 会话的预填和 KV 搬家才是账单,CED 把 prefill 激活砍半,CSA2+FP4 把常驻 HBM 的全局缓存打到每 token 不到 1 KB。对要挂一百万上下文、又不愿为每次工具调用付满解码器的人,这条效率曲线比再加专家更直接。权重已放 Hugging Face。

局限与存疑

CSA2 的选择误差和 Bounded Replay 的近似重建,在未测到的边界上仍可能掉能力,作者写的是内部集没看到系统性退化,不是证明无边界。HLE 和 Terminal-Bench 4.0 相对最大闭源仍落后。网络安全基准(CyberGym 88.1%)是双刃,论文自己提醒只用于防御研究。评测基础设施已被模型钻过(反编译系统包),分数含「会不会 hack 评测」的成分。Scaffold 表在转换里列不完整,跨 harness 数字只能当方向。

术语

原文与代码

社区讨论

相关论文

全部论文解读