{"id":113787,"date":"2026-10-07T12:49:51","date_gmt":"2026-10-07T04:49:51","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/113787.html"},"modified":"2026-10-07T12:49:51","modified_gmt":"2026-10-07T04:49:51","slug":"%e3%80%90cs336%e3%80%91lecture4-%e7%ba%bf%e6%80%a7%e6%b3%a8%e6%84%8f%e5%8a%9bmambagdndsa","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/113787.html","title":{"rendered":"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA"},"content":{"rendered":"

\u56de\u987e\u4f20\u7edf\u6ce8\u610f\u529b<\/h2>\n

\u8fd9\u8282\u8bb2\u7684\u662f\u6ce8\u610f\u529b\u53d8\u4f53，\u5148\u56de\u987e\u4e00\u4e0b\u4f20\u7edf\u6ce8\u610f\u529b<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u8ba1\u7b97\u5f00\u9500\u968f\u4e0a\u4e0b\u6587\u7a97\u53e3\u662f\u5e73\u65b9\u91cf\u7ea7\u589e\u957f\u7684，\u4e5f\u5c31\u662fO(n)O(n)<\/span><\/span>O<\/span>(<\/span>n<\/span>)<\/span><\/span><\/span><\/span><\/span>\u4e0a\u4e0b\u6587，O(n2)O(n^2)<\/span><\/span>O<\/span>(<\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u7684\u8ba1\u7b97\u590d\u6742\u5ea6，\u5982\u53f3\u56fe\u3002\u4f46\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3\u53c8\u662f\u5f88\u91cd\u8981\u7684\u9700\u6c42，\u73b0\u5728\u6a21\u578b\u4e3b\u6d41\u7684\u8fdb\u5316\u65b9\u5411\u4e4b\u4e00\u5c31\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3，\u5982\u5de6\u56fe\u3002<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u4e0a\u4e00\u8bb2，\u5bf9\u6b64\u7684\u89e3\u51b3\u65b9\u5f0f\u662f\u5f15\u5165\u6df7\u5408\u6ce8\u610f\u529b，\u7a00\u758f\u6ce8\u610f\u529b\u3002\u5982\u5de6\u56fe，\u5c55\u793a\u4e86\u4e00\u5c42\u6ce8\u610f\u529b\u6a21\u5757\u7684\u5c42\u5185\u7a00\u758f，\u4ee5\u53ca\u5c42\u95f4\u7a00\u758f，\u4e5f\u5c31\u662f\u51e0\u5c42\u7a00\u758f\u6ce8\u610f\u529b，\u4e00\u5c42\u5b8c\u6574\u6ce8\u610f\u529b\u3002\u4f46\u8fd9\u6837\u7684\u65b9\u5f0f\u4f18\u5316\u6548\u679c\u8fd8\u662f\u4e0d\u591f\u597d\u3002<\/p>\n

\u53e6\u4e00\u79cd\u4f18\u5316\u601d\u8def\u662fflashattention\u4e3a\u4ee3\u8868\u7684\u7cfb\u7edf\u7ea7\u4f18\u5316，\u5982\u53f3\u56fe，\u76f8\u6bd4\u6734\u7d20pytorch\u541e\u5410\u6709\u63d0\u9ad8，\u4f46\u5728\u4e0a\u4e0b\u6587\u7a97\u53e3\u62c9\u957f\u65f6\u4ecd\u7136\u4f1a\u9047\u5230\u74f6\u9888，\u56fe\u4e2d\u5c55\u793a\u7684\u5c31\u662f\u4e0a\u4e0b\u6587\u7a97\u53e3\u589e\u5927\u65f6，\u6ce8\u610f\u529b\u5c42\u6bcf\u79d2\u541e\u5410\u91cf\u7684\u53d8\u5316，\u53ef\u4ee5\u770b\u5230\u57284k\u540e\u541e\u5410\u5c31\u51e0\u4e4e\u4e0d\u589e\u957f\u4e86\u3002<\/p>\n

\u56e0\u6b64\u6b63\u5982\u6700\u4e0b\u65b9\u6587\u5b57\u6240\u8bf4\u7684，\u6211\u4eec\u8981\u4e00\u79cd\u66f4\u6fc0\u8fdb\u7684\u4f18\u5316，\u8fd9\u5c31\u662f\u672c\u8282\u524d\u534a\u90e8\u5206\u7684\u4e3b\u9898：\u7ebf\u6027\u6ce8\u610f\u529b<\/p>\n

\u6734\u7d20\u7ebf\u6027\u6ce8\u610f\u529b<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u6700\u6734\u7d20\u7684\u7ebf\u6027\u6ce8\u610f\u529b\u6765\u6e90\u4e8e\u4e00\u4e2a\u7b80\u5355\u7684\u60f3\u6cd5，\u6ce8\u610f\u539f\u59cb\u6ce8\u610f\u529b\u516c\u5f0f，\u5982\u679c\u5ffd\u7565\u90a3\u4e2asoftmax\u51fd\u6570，\u5c31\u662f\u4e09\u4e2a\u77e9\u9635\u8fde\u4e58\u3002\u77e9\u9635\u8fde\u4e58，\u8c03\u6574\u7ed3\u5408\u4f18\u5148\u7ea7\u53ef\u4ee5\u964d\u4f4e\u8ba1\u7b97\u91cf，\u8fd9\u662f\u4e2a\u533a\u95f4DP\u7684\u7ecf\u5178\u95ee\u9898\u3002\u5728\u8fd9\u91cc\u53ea\u6709\u4e09\u4e2a\u77e9\u9635，\u5b9e\u9645\u4e0a\u53ea\u6709\u4e24\u79cd\u7ed3\u5408\u65b9\u6cd5，QK\u5148\u7ed3\u5408，KV\u5148\u7ed3\u5408\u3002\u6807\u51c6\u6ce8\u610f\u529b\u662fQK\u7ed3\u5408，\u8fd9\u91cc\u6362\u6210KV\u5148\u7ed3\u5408\u80fd\u5426\u51cf\u5c0f\u8ba1\u7b97\u91cf？<\/p>\n

\u6765\u5206\u6790\u4e00\u4e0b\u590d\u6742\u5ea6，QKV\u90fd\u662f(n,d)\u7684\u3002\u6807\u51c6\u6ce8\u610f\u529b\u7684QK\u76f8\u4e58，\u590d\u6742\u5ea6n2dn^2d<\/span><\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>d<\/span><\/span><\/span><\/span><\/span>，\u6ce8\u610f\u529b\u5f97\u5206\u662f(n,n)\u7684，\u518d\u548cV\u76f8\u4e58，\u4e5f\u662fn2dn^2d<\/span><\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>d<\/span><\/span><\/span><\/span><\/span>\u7684\u3002\u5982\u679c\u5148\u7ed3\u5408KV，\u6ce8\u610fK\u4f9d\u65e7\u662f\u8f6c\u7f6e\u7684，\u8ba1\u7b97\u91cfnd2nd^2<\/span><\/span>n<\/span>d<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u5f97\u5230\u7684\u7ed3\u679c\u662f(d,d)\u7684，\u518d\u548cQ\u76f8\u4e58，\u8ba1\u7b97\u91cf\u4e5f\u662fnd2nd^2<\/span><\/span>n<\/span>d<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

\u6ce8\u610f\u5230\u5728\u6a21\u578b\u67b6\u6784\u4e2d，\u9690\u85cf\u5c42\u7ef4\u5ea6d\u4e0d\u53d8，\u4e14\u4e00\u822c\u76f8\u5bf9\u4e8e\u4e0a\u4e0b\u6587\u957f\u5ea6n\u6765\u8bf4\u662f\u5f88\u5c0f\u7684，\u53ef\u4ee5\u89c6\u4e3a\u5e38\u6570\u3002\u4e5f\u5c31\u662f\u8bf4\u8fd9\u4e2a\u6539\u53d8\u8ba1\u7b97\u987a\u5e8f\u7684\u7b80\u5355\u4f18\u5316\u5c31\u80fd\u628a\u8ba1\u7b97\u590d\u6742\u5ea6\u4eceO(n2)O(n^2)<\/span><\/span>O<\/span>(<\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u4f18\u5316\u5230O(n)O(n)<\/span><\/span>O<\/span>(<\/span>n<\/span>)<\/span><\/span><\/span><\/span><\/span>！<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u57fa\u4e8e\u8fd9\u4e2a\u5f62\u5f0f，\u53ef\u4ee5\u628a\u6a21\u578b\u7b49\u4ef7\u8f6c\u6362\u6210\u4e00\u4e2a\u7c7bRNN\u7684\u5f62\u5f0f，\u6bcf\u4e2atoken\u8ba1\u7b97ktvtTk_tv_t^T<\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>v<\/span><\/span>t<\/span><\/span><\/span><\/span>T<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u5f97\u5230\u5f53\u524dtoken\u7684(d,d)\u72b6\u6001，\u7d2f\u52a0\u5230\u9690\u85cf\u5c42\u72b6\u6001StS_t<\/span><\/span>S<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u4e0a\u3002\u7136\u540e\u5229\u7528\u9690\u85cf\u5c42\u72b6\u6001StS_t<\/span><\/span>S<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u548cqtq_t<\/span><\/span>q<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u8ba1\u7b97\u8f93\u51fatoken，\u8fd9\u91ccqtq_t<\/span><\/span>q<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u5c31\u7c7b\u4f3cRNN\u7684\u8f93\u51fa\u95e8\u63a7，kt,vtk_t,v_t<\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>,<\/span><\/span>v<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u7c7b\u4f3c\u4e8eRNN\u7684\u8f93\u5165\u95e8\u63a7\u548c\u66f4\u65b0\u95e8\u63a7\u3002<\/p>\n

\u8fd9\u6837\u8ba1\u7b97\u662f\u4e32\u884c\u7684，\u4f46\u662f\u63a8\u7406\u65f6decode\u672c\u6765\u5c31\u662f\u4e32\u884c\u7684，\u4f46\u901a\u8fc7\u7ebf\u6027\u6ce8\u610f\u529b，\u6211\u4eec\u4ecd\u7136\u964d\u4f4e\u4e86\u63a8\u7406\u7684\u8ba1\u7b97\u91cf，\u539f\u6765\u6bcf\u4e2atoken\u8981\u548c\u524d\u9762n\u4e2atoken\u8ba1\u7b97\u6ce8\u610f\u529b，\u505a\u4e00\u4e2agemv\u64cd\u4f5c，\u6bcf\u4e2atoken\u662fO(nd)O(nd)<\/span><\/span>O<\/span>(<\/span>n<\/span>d<\/span>)<\/span><\/span><\/span><\/span><\/span>\u7684\u590d\u6742\u5ea6，\u73b0\u5728\u53ea\u7528\u81ea\u5df1\u505a\u4e00\u4e2aktvtTk_tv_t^T<\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>v<\/span><\/span>t<\/span><\/span><\/span><\/span>T<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u662f\u4e24\u4e2a(1,d)\u7684\u5411\u91cf\u5916\u79ef，\u590d\u6742\u5ea6O(d2)O(d^2)<\/span><\/span>O<\/span>(<\/span>d<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u3002\u540c\u6837\u63a8\u7406\u4e00\u6bb5\u957fn\u7684\u6587\u672c，\u6807\u51c6\u6ce8\u610f\u529b\u590d\u6742\u5ea6O(n2d)O(n^2d)<\/span><\/span>O<\/span>(<\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>d<\/span>)<\/span><\/span><\/span><\/span><\/span>，\u8fd9\u4e2a\u7c7bRNN\u7684\u590d\u6742\u5ea6O(nd2)O(nd^2)<\/span><\/span>O<\/span>(<\/span>n<\/span>d<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>，\u548c\u54b1\u4eec\u524d\u9762\u5206\u6790\u4e00\u81f4，\u8fd9\u91cc\u53ea\u4e0d\u8fc7\u662f\u4ece\u63a8\u7406decode\u89d2\u5ea6\u5206\u6790\u4e86\u4e00\u904d\u3002<\/p>\n

\u7531\u4e8e\u5728\u6570\u5b66\u4e0a\u600e\u4e48\u7ed3\u5408\u90fd\u662f\u7b49\u4ef7\u7684，\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u5728\u8bad\u7ec3\u65f6\u4ecd\u7528\u539f\u6765\u7684\u6ce8\u610f\u529b\u516c\u5f0f，\u5b9e\u73b0\u63a9\u7801+\u81ea\u56de\u5f52\u8bad\u7ec3\u7684\u5e76\u884c\u8bad\u7ec3\u3002\u5728\u63a8\u7406\u65f6\u624d\u7528\u8fd9\u4e2aRNN\u5f62\u5f0f\u3002\u8fd9\u6837\u53ef\u4ee5\u540c\u65f6\u83b7\u5f97\u8bad\u7ec3\u548c\u63a8\u7406\u65f6\u7684\u52a0\u901f\u3002<\/p>\n

\u4e3e\u4f8b:Minimax<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\nMinimax\u5c31\u91c7\u7528\u4e86\u7ebf\u6027\u6ce8\u610f\u529b，\u51c6\u786e\u6765\u8bf4\u662f\u7ebf\u6027\u6ce8\u610f\u529b\u548c\u6807\u51c6\u6ce8\u610f\u529b\u7684\u6df7\u5408\u67b6\u6784，7\u5c42\u7ebf\u6027，1\u5c42\u6807\u51c6\u3002\u6a21\u578b\u80fd\u529b\u4e0d\u8f93\u5f53\u65f6\u5927\u90e8\u5206\u4e3b\u6d41\u6a21\u578b，\u89c1\u5de6\u56fe，\u53ef\u89c1\u7ebf\u6027\u6ce8\u610f\u529b\u6ca1\u6709\u8fc7\u591a\u7684\u635f\u5931\u6027\u80fd\u3002\u4f46\u8ba1\u7b97\u91cf\u5927\u5e45\u964d\u4f4e，\u89c1\u4e2d\u56fe，\u5176\u4ed6\u6a21\u578b\u7684\u8ba1\u7b97\u91cf，\u968f\u4e0a\u4e0b\u6587\u589e\u957f\u662f\u5e73\u65b9\u91cf\u7ea7\u589e\u957f\u7684，\u800cMinimax-M1\u51e0\u4e4e\u662f\u7ebf\u6027\u589e\u957f\u7684，\u8fd9\u5c31\u662f\u7ebf\u6027\u6ce8\u610f\u529b\u7684\u529f\u52b3\u3002<\/p>\n

\u7ee7\u7eed\u6539\u8fdb：Mamba<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u7ed9\u524d\u9762\u7684\u6734\u7d20\u7ebf\u6027\u6ce8\u610f\u529b\u7684\u72b6\u6001\u66f4\u65b0\u65f6\u52a0\u4e0a\u4e86\u95e8\u63a7\u03b3，\u95e8\u63a7\u76f4\u63a5\u662f\u4e58\u4e0a\u4e00\u4e2a\u5e38\u6570，\u8ba1\u7b97\u91cf\u53ea\u589e\u52a0\u4e86\u5e38\u6570\u7ea7，\u4f46\u662f\u95e8\u63a7\u80fd\u6709\u6548\u63d0\u5347\u6a21\u578b\u80fd\u529b\u3002\u4ee5\u53ca\u8f93\u51fa\u65f6\u589e\u52a0\u4e86\u4e00\u4e2a\u542bv\u7684\u9879，\u8fd9\u7c7b\u4f3c\u4e8eResNet\u7684\u6b8b\u5dee\u8fde\u63a5\u9879，\u53ef\u4ee5\u8ba9\u8f93\u5165\u76f4\u63a5\u5f71\u54cd\u5230\u8f93\u51fa，\u4e0d\u7ecf\u8fc7\u9690\u85cf\u5c42\u7684\u538b\u7f29，\u4e5f\u662f\u53ef\u4ee5\u63d0\u5347\u6a21\u578b\u80fd\u529b，\u540c\u65f6\u8ba1\u7b97\u91cf\u4e5f\u53ea\u589e\u52a0\u4e86\u5e38\u6570\u7ea7\u3002<\/p>\n

\u4e3e\u4f8b：Nemotron<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u4f7f\u7528Mamba\u7684\u4f8b\u5b50\u662f\u82f1\u4f1f\u8fbe\u5f00\u6e90\u7684Nemotron3，\u91c7\u7528\u7684\u662f3\u5c42mamba，\u4e00\u5c42\u6807\u51c6\u6ce8\u610f\u529b\u7684\u6df7\u5408\u67b6\u6784<\/p>\n

\u7ee7\u7eed\u6539\u8fdb：GDN<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u5728mamba\u7684\u57fa\u7840\u4e0a，Gated Delta Net，\u5f15\u5165\u4e86delta\u89c4\u5219，\u540c\u65f6\u8fd8\u4fdd\u7559\u4e86\u95e8\u63a7，\u8fd9\u4e5f\u662f\u4ed6\u540d\u5b57\u91ccGated\u548cDelta\u7684\u6765\u5386\u3002\u6838\u5fc3\u601d\u60f3\u662f\u5148\u4e58\u4e0a\u4e00\u4e2a(It\u2212\u03b2tktktT)(I_t-\\\\beta_tk_tk_t^T)<\/span><\/span>(<\/span>I<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2212<\/span><\/span><\/span><\/span>\u03b2<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>T<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u9879\u64e6\u9664\u90e8\u5206\u8bb0\u5fc6，\u518d\u52a0\u5165\u65b0token\u66f4\u65b0\u8bb0\u5fc6，\u8fd9\u4f7f\u5f97\u6a21\u578b\u8868\u8fbe\u80fd\u529b\u8fdb\u4e00\u6b65\u589e\u5f3a\u3002\u770b\u8d77\u6765\u6709\u4e24\u4e2a(d,d)\u7684\u77e9\u9635\u76f8\u4e58，\u5355token\u8ba1\u7b97\u91cf\u589e\u957f\u5230O(d3)O(d^3)<\/span><\/span>O<\/span>(<\/span>d<\/span><\/span>3<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>，\u4f46\u5b9e\u9645\u4e0a，\u4ecd\u7136\u53ef\u4ee5\u7528\u6211\u4eec\u4e4b\u524d\u63d0\u51fa\u7ebf\u6027\u6ce8\u610f\u529b\u65f6\u7684\u6280\u5de7，\u6539\u53d8\u77e9\u9635\u4e58\u6cd5\u987a\u5e8f，\u5148\u505aktTst\u22121k_t^Ts_{t-1}<\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>T<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>s<\/span><\/span>t<\/span>\u2212<\/span>1<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u518d\u4e58\u4e0aktk_t<\/span><\/span>k<\/span><\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u6574\u4f53\u8ba1\u7b97\u91cf\u8fd8\u662fO(d2)O(d^2)<\/span><\/span>O<\/span>(<\/span>d<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u7684，\u53ea\u662f\u5e38\u6570\u5927\u4e00\u70b9\u3002<\/p>\n

\u4e3e\u4f8b：Qwen<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u4f7f\u7528GDN\u6700\u7ecf\u5178\u7684\u4f8b\u5b50\u662fQwen\u3002\u5982\u53f3\u56fe，，\u4f7f\u7528GDN\u7684Qwen Next，\u76f8\u6bd4\u4f7f\u7528\u6807\u51c6\u6ce8\u610f\u529b\u7684Qwen3，\u5f53\u4e0a\u4e0b\u6587\u7a97\u53e3\u589e\u957f\u65f6，\u541e\u5410\u4e5f\u968f\u4e4b\u7ebf\u6027\u589e\u957f，\u800c\u6807\u51c6\u6ce8\u610f\u529b\u7684\u541e\u5410\u51e0\u4e4e\u4e0d\u53d8，\u751a\u81f3\u8fd8\u7565\u6709\u964d\u4f4e\u3002<\/p>\n

\u6d88\u878d\u5b9e\u9a8c<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u6d88\u878d\u5b9e\u9a8c\u8bc1\u660e\u968f\u7740\u7ebf\u6027\u6ce8\u610f\u529b：\u6807\u51c6\u6ce8\u610f\u529b\u7684\u6bd4\u4f8b\u63d0\u5347，\u6a21\u578b\u8868\u8fbe\u80fd\u529b\u786e\u5b9e\u4f1a\u4e0b\u964d，\u4e5f\u5c31\u662f\u7ebf\u6027\u6ce8\u610f\u529b\u786e\u5b9e\u4f1a\u635f\u5931\u8868\u8fbe\u80fd\u529b\u3002\u4f46\u662f\u7ebf\u6027\u6ce8\u610f\u529b\u6bd4\u4f8b\u8f83\u4f4e\u65f6，\u6a21\u578b\u80fd\u529b\u635f\u5931\u7684\u5e76\u4e0d\u591a，\u53ef\u4ee5\u540c\u65f6\u83b7\u5f97\u8f83\u597d\u7684\u6a21\u578b\u80fd\u529b，\u548c\u7ebf\u6027\u6ce8\u610f\u529b\u7684\u9ad8\u541e\u5410\u3002<\/p>\n

\u53e6\u4e00\u79cd\u6ce8\u610f\u529b\u53d8\u4f53：DSA<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\n\u53e6\u4e00\u79cd\u964d\u4f4e\u6ce8\u610f\u529b\u590d\u6742\u5ea6\u7684\u65b9\u6cd5\u662fdeepseek\u63d0\u51fa\u7684deepseek sparse attention，\u6838\u5fc3\u601d\u8def\u662f\u5148\u7528\u4e00\u4e2a\u8f7b\u91cf\u7ea7\u7684\u7d22\u5f15\u5934indexer，\u8ba1\u7b97\u51fa\u5f53\u524dtoken\u548c\u524d\u9762\u6240\u6709token\u7684\u76f8\u5173\u6027\u5f97\u5206，\u7136\u540e\u9009\u51fatopk\u7684token\u4e0e\u5f53\u524dtoken\u8ba1\u7b97\u6ce8\u610f\u529b\u3002<\/p>\n

\u4f18\u52bf\u5728\u4e8e：\u7d22\u5f15\u5934\u662f\u5f88\u8f7b\u7684，\u8ba1\u7b97\u5f00\u9500\u5f88\u4f4e，\u63a5\u4e0b\u6765\u53ea\u9009topk，\u628a\u8ba1\u7b97\u91cf\u4eceO(n2)O(n^2)<\/span><\/span>O<\/span>(<\/span>n<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>\u964d\u4f4e\u5230O(nk)O(nk)<\/span><\/span>O<\/span>(<\/span>nk<\/span>)<\/span><\/span><\/span><\/span><\/span>，\u5e76\u4e14\u4e00\u822ck\u8fdc\u5c0f\u4e8en，\u4e14k\u4e0d\u968fn\u589e\u957f\u800c\u589e\u5927，\u56e0\u6b64\u53ef\u4ee5\u89c6\u4e3a\u5e38\u6570，\u56e0\u6b64DSA\u4e5f\u53ef\u4ee5\u89c6\u4e3aO(n)O(n)<\/span><\/span>O<\/span>(<\/span>n<\/span>)<\/span><\/span><\/span><\/span><\/span>\u590d\u6742\u5ea6，\u4e5f\u662f\u4e00\u79cd\u7ebf\u6027\u6ce8\u610f\u529b\u3002<\/p>\n

\u5e76\u4e14\u7d22\u5f15\u5934\u662f\u53ef\u4ee5\u5355\u72ec\u8bad\u7ec3\u7684，\u4e5f\u5c31\u662f\u53ef\u4ee5\u5728\u4e00\u4e2a\u8bad\u7ec3\u597d\u7684\u6807\u51c6\u6ce8\u610f\u529b\u5c42\u4e0a，\u63a5\u4e0a\u4e00\u4e2aindexer，\u5355\u72ec\u8bad\u7ec3，\u53ea\u9700\u8981\u4e00\u70b9\u5fae\u8c03\u8bad\u7ec3\u5373\u53ef\u5b9e\u73b0DSA\u3002\u8bad\u7ec3\u6210\u672c\u4e5f\u5f88\u4f4e\u3002<\/p>\n

\u4e3e\u4f8b：DeepSeek V3.2\/GLM5<\/h2>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"
\nGLM5\u548cDS V3.2\u90fd\u7528\u4e86DSA\u67b6\u6784，\u6a21\u578b\u80fd\u90fd\u4e0d\u9519\u3002\u540c\u65f6\u6ce8\u610f\u5de6\u4e0b，\u4f7f\u7528DSA\u76843.2\u76f8\u6bd4\u4e0d\u4f7f\u7528\u7684\u524d\u4ee33.1，\u63a8\u7406\u6210\u672c\u5927\u5e45\u4e0b\u964d，\u968f\u7740\u4e0a\u4e0b\u6587\u589e\u957f，\u6bcf\u767e\u4e07token\u7684\u63a8\u7406\u6210\u672c\u51e0\u4e4e\u8fd8\u662f\u4e00\u6761\u76f4\u7ebf，\u659c\u7387\u4f4e\u7684\u53ef\u6015\u3002\u8fd9\u4e5f\u662fdeepseek api\u4ef7\u683c\u8fdc\u4f4e\u4e8e\u5176\u4ed6\u516c\u53f8\u7684\u539f\u56e0\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u56de\u987e\u4f20\u7edf\u6ce8\u610f\u529b
\n\u8fd9\u8282\u8bb2\u7684\u662f\u6ce8\u610f\u529b\u53d8\u4f53，\u5148\u56de\u987e\u4e00\u4e0b\u4f20\u7edf\u6ce8\u610f\u529b\u8ba1\u7b97\u5f00\u9500\u968f\u4e0a\u4e0b\u6587\u7a97\u53e3\u662f\u5e73\u65b9\u91cf\u7ea7\u589e\u957f\u7684，\u4e5f\u5c31\u662fO(n)O(n)O(n)\u4e0a\u4e0b\u6587，O(n2)O(n^2)O(n2)\u7684\u8ba1\u7b97\u590d\u6742\u5ea6，\u5982\u53f3\u56fe\u3002\u4f46\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3\u53c8\u662f\u5f88\u91cd\u8981\u7684\u9700\u6c42，\u73b0\u5728\u6a21\u578b\u4e3b\u6d41\u7684\u8fdb\u5316\u65b9\u5411\u4e4b\u4e00\u5c31\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3，\u5982\u5de6\u56fe\u3002\u4e0a\u4e00\u8bb2，\u5bf9\u6b64\u7684\u89e3\u51b3\u65b9\u5f0f\u662f\u5f15\u5165\u6df7\u5408\u6ce8\u610f\u529b，\u7a00\u758f\u6ce8\u610f\u529b\u3002\u5982\u5de6\u56fe，\u5c55\u793a\u4e86\u4e00\u5c42\u6ce8\u610f\u529b<\/p>\n","protected":false},"author":2,"featured_media":113775,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[8751,12947,12945,11653,12946,3035],"topic":[],"class_list":["post-113787","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-cs336","tag-dsa","tag-gdn","tag-11653","tag-12946","tag-mamba"],"yoast_head":"\n\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/113787.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u56de\u987e\u4f20\u7edf\u6ce8\u610f\u529b \u8fd9\u8282\u8bb2\u7684\u662f\u6ce8\u610f\u529b\u53d8\u4f53，\u5148\u56de\u987e\u4e00\u4e0b\u4f20\u7edf\u6ce8\u610f\u529b\u8ba1\u7b97\u5f00\u9500\u968f\u4e0a\u4e0b\u6587\u7a97\u53e3\u662f\u5e73\u65b9\u91cf\u7ea7\u589e\u957f\u7684，\u4e5f\u5c31\u662fO(n)O(n)O(n)\u4e0a\u4e0b\u6587，O(n2)O(n^2)O(n2)\u7684\u8ba1\u7b97\u590d\u6742\u5ea6，\u5982\u53f3\u56fe\u3002\u4f46\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3\u53c8\u662f\u5f88\u91cd\u8981\u7684\u9700\u6c42，\u73b0\u5728\u6a21\u578b\u4e3b\u6d41\u7684\u8fdb\u5316\u65b9\u5411\u4e4b\u4e00\u5c31\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3，\u5982\u5de6\u56fe\u3002\u4e0a\u4e00\u8bb2，\u5bf9\u6b64\u7684\u89e3\u51b3\u65b9\u5f0f\u662f\u5f15\u5165\u6df7\u5408\u6ce8\u610f\u529b，\u7a00\u758f\u6ce8\u610f\u529b\u3002\u5982\u5de6\u56fe，\u5c55\u793a\u4e86\u4e00\u5c42\u6ce8\u610f\u529b\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/113787.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-07T04:49:51+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007044939-6ac5cf633366f.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113787.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/113787.html\",\"name\":\"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-07T04:49:51+00:00\",\"dateModified\":\"2026-10-07T04:49:51+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/113787.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/113787.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113787.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/113787.html","og_locale":"zh_CN","og_type":"article","og_title":"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u56de\u987e\u4f20\u7edf\u6ce8\u610f\u529b \u8fd9\u8282\u8bb2\u7684\u662f\u6ce8\u610f\u529b\u53d8\u4f53，\u5148\u56de\u987e\u4e00\u4e0b\u4f20\u7edf\u6ce8\u610f\u529b\u8ba1\u7b97\u5f00\u9500\u968f\u4e0a\u4e0b\u6587\u7a97\u53e3\u662f\u5e73\u65b9\u91cf\u7ea7\u589e\u957f\u7684，\u4e5f\u5c31\u662fO(n)O(n)O(n)\u4e0a\u4e0b\u6587，O(n2)O(n^2)O(n2)\u7684\u8ba1\u7b97\u590d\u6742\u5ea6，\u5982\u53f3\u56fe\u3002\u4f46\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3\u53c8\u662f\u5f88\u91cd\u8981\u7684\u9700\u6c42，\u73b0\u5728\u6a21\u578b\u4e3b\u6d41\u7684\u8fdb\u5316\u65b9\u5411\u4e4b\u4e00\u5c31\u662f\u62d3\u5c55\u4e0a\u4e0b\u6587\u7a97\u53e3，\u5982\u5de6\u56fe\u3002\u4e0a\u4e00\u8bb2，\u5bf9\u6b64\u7684\u89e3\u51b3\u65b9\u5f0f\u662f\u5f15\u5165\u6df7\u5408\u6ce8\u610f\u529b，\u7a00\u758f\u6ce8\u610f\u529b\u3002\u5982\u5de6\u56fe，\u5c55\u793a\u4e86\u4e00\u5c42\u6ce8\u610f\u529b","og_url":"https:\/\/www.wsisp.com\/helps\/113787.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-07T04:49:51+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007044939-6ac5cf633366f.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"3 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/113787.html","url":"https:\/\/www.wsisp.com\/helps\/113787.html","name":"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-07T04:49:51+00:00","dateModified":"2026-10-07T04:49:51+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/113787.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/113787.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/113787.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u3010CS336\u3011lecture4 \u7ebf\u6027\u6ce8\u610f\u529b|Mamba|GDN|DSA"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/113787","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=113787"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/113787\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/113775"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=113787"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=113787"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=113787"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=113787"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}