{"id":5660,"date":"2025-02-19T17:31:14","date_gmt":"2025-02-19T09:31:14","guid":{"rendered":"https:\/\/nullthought.net\/?p=5660"},"modified":"2025-02-23T11:08:03","modified_gmt":"2025-02-23T03:08:03","slug":"ultramem%ef%bc%9a%e5%88%a9%e7%94%a8%e5%a4%a7%e8%a7%84%e6%a8%a1%e7%9a%84%e8%b6%85%e7%a8%80%e7%96%8f%e5%86%85%e5%ad%98%e5%b1%82%ef%bc%88ultra-sparse-memory-layer%ef%bc%89%e6%98%be%e8%91%97%e6%8f%90","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=5660","title":{"rendered":"UltraMem\uff1a\u5229\u7528\u5927\u89c4\u6a21\u7684\u8d85\u7a00\u758f\u5185\u5b58\u5c42\uff08Ultra-Sparse Memory Layer\uff09\u663e\u8457\u63d0\u5347Transformer\u6a21\u578b\u6027\u80fd"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2411.12364\" target=\"_blank\" rel=\"noreferrer noopener\">Ultra-Sparse Memory Network<\/a><\/strong>\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3aUltraMem\u7684\u65b0\u578b\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c\u89e3\u51b3\u4f20\u7edfTransformer\u6a21\u578b\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u56e0\u5185\u5b58\u8bbf\u95ee\u9ad8\u5f00\u9500\u5bfc\u81f4\u7684\u6027\u80fd\u74f6\u9888\u95ee\u9898\u3002UltraMem\u901a\u8fc7\u5f15\u5165\u5927\u89c4\u6a21\u7684\u8d85\u7a00\u758f\u5185\u5b58\u5c42\uff08Ultra-Sparse Memory Layer\uff09\uff0c\u663e\u8457\u63d0\u9ad8\u4e86\u8ba1\u7b97\u6548\u7387\uff0c\u51cf\u5c11\u4e86\u63a8\u7406\u5ef6\u8fdf\uff0c\u540c\u65f6\u4fdd\u6301\u6216\u63d0\u9ad8\u4e86\u6a21\u578b\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aZihao Huang, Qiyang Min, Hongzhi Huang, Defa Zhu, Yutao Zeng, Ran Guo, Xun Zhou\uff0c\u5747\u6765\u81eaBytedance\uff08\u5b57\u8282\u8df3\u52a8\uff09\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"498\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-12-1024x498.png\" alt=\"\" class=\"wp-image-5661\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-12-1024x498.png 1024w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-12-300x146.png 300w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-12-768x373.png 768w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-12.png 1029w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><strong><a href=\"https:\/\/arxiv.org\/abs\/2411.12364\" target=\"_blank\" rel=\"noreferrer noopener\">Ultra-Sparse Memory Network<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e00\u3001\u5f15\u8a00<\/h5>\n\n\n\n<p>\u8fd1\u5e74\u6765\uff0c\u968f\u7740\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP\uff09\u9886\u57df\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u5feb\u901f\u53d1\u5c55\uff0c\u6a21\u578b\u7684\u8ba1\u7b97\u9700\u6c42\u4e5f\u5448\u6307\u6570\u589e\u957f\u3002\u5927\u89c4\u6a21\u7684Transformer\u6a21\u578b\uff08\u5982GPT-3\u7b49\uff09\u867d\u7136\u5728\u5404\u79cd\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u5b83\u4eec\u7684\u63a8\u7406\u8fc7\u7a0b\u5f80\u5f80\u53d7\u5230\u8ba1\u7b97\u8d44\u6e90\u548c\u5185\u5b58\u5e26\u5bbd\u7684\u9650\u5236\uff0c\u5bfc\u81f4\u63a8\u7406\u901f\u5ea6\u663e\u8457\u4e0b\u964d\u3002\u4e3a\u4e86\u63d0\u9ad8\u8ba1\u7b97\u6548\u7387\uff0c\u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e9b\u65b9\u6cd5\uff0c\u4f8b\u5982Mixture of Experts\uff08MoE\uff09\u548cProduct Key Memory\uff08PKM\uff09\u3002MoE\u901a\u8fc7\u9009\u62e9\u6027\u6fc0\u6d3b\u4e0d\u540c\u7684\u5b50\u6a21\u578b\uff08\u5373\u201c\u4e13\u5bb6\u201d\uff09\uff0c\u4ece\u800c\u589e\u52a0\u6a21\u578b\u7684\u53c2\u6570\u6570\u91cf\u800c\u4e0d\u663e\u8457\u589e\u52a0\u8ba1\u7b97\u5f00\u9500\u3002\u7136\u800c\uff0c\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\uff0cMoE\u4ecd\u7136\u5b58\u5728\u8f83\u9ad8\u7684\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\uff0c\u56e0\u4e3a\u5373\u4f7f\u6fc0\u6d3b\u4e86\u90e8\u5206\u4e13\u5bb6\uff0c\u4ecd\u7136\u9700\u8981\u5927\u91cf\u7684\u5185\u5b58\u8bbf\u95ee\u4ee5\u5b8c\u6210\u8ba1\u7b97\u3002<\/p>\n\n\n\n<p>PKM\u65b9\u6cd5\u5219\u901a\u8fc7\u5f15\u5165\u5916\u90e8\u5185\u5b58\u6a21\u5757\uff0c\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u7684\u9891\u7387\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u8ba1\u7b97\u6548\u7387\uff0c\u4f46\u5176\u6027\u80fd\u4ecd\u4e0d\u5982MoE\u3002\u4e3a\u4e86\u89e3\u51b3MoE\u548cPKM\u7684\u4e0d\u8db3\uff0cUltraMem\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u67b6\u6784\uff0c\u901a\u8fc7\u5f15\u5165\u5927\u89c4\u6a21\u3001\u8d85\u7a00\u758f\u7684\u5185\u5b58\u5c42\uff0c\u5728\u4fdd\u8bc1\u8ba1\u7b97\u6548\u7387\u7684\u540c\u65f6\uff0c\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u5ef6\u8fdf\u3002\u4e0eMoE\u76f8\u6bd4\uff0cUltraMem\u5728\u63a8\u7406\u901f\u5ea6\u4e0a\u6709\u663e\u8457\u63d0\u9ad8\uff0c\u5e76\u4e14\u80fd\u591f\u5728\u76f8\u540c\u8ba1\u7b97\u8d44\u6e90\u4e0b\u8fbe\u5230\u66f4\u597d\u7684\u6a21\u578b\u6027\u80fd\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e8c\u3001\u76f8\u5173\u5de5\u4f5c<\/h5>\n\n\n\n<p>\u8bba\u6587\u9996\u5148\u56de\u987e\u4e86\u4e0eUltraMem\u76f8\u5173\u7684\u7814\u7a76\u5de5\u4f5c\uff0c\u4e3b\u8981\u96c6\u4e2d\u5728Mixture of Experts\uff08MoE\uff09\u548cProduct Key Memory\uff08PKM\uff09\u4e24\u79cd\u65b9\u6cd5\u3002MoE\uff08Shazeer et al., 2017\uff09\u6700\u65e9\u63d0\u51fa\uff0c\u901a\u8fc7\u5728\u6bcf\u4e2a\u63a8\u7406\u6b65\u9aa4\u4e2d\u6fc0\u6d3b\u4e00\u90e8\u5206\u201c\u4e13\u5bb6\u201d\u6a21\u578b\uff0c\u6765\u589e\u52a0\u6a21\u578b\u7684\u53c2\u6570\u6570\u91cf\uff0c\u540c\u65f6\u63a7\u5236\u8ba1\u7b97\u5f00\u9500\u3002\u5c3d\u7ba1MoE\u53ef\u4ee5\u5728\u8bad\u7ec3\u65f6\u6709\u6548\u5229\u7528\u5927\u91cf\u53c2\u6570\uff0c\u4f46\u7531\u4e8e\u63a8\u7406\u8fc7\u7a0b\u4e2d\u9700\u8981\u8bbf\u95ee\u5927\u91cf\u5185\u5b58\uff0c\u5bfc\u81f4\u5176\u63a8\u7406\u901f\u5ea6\u8f83\u6162\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0cFedus et al.\uff082022\uff09\u8fdb\u4e00\u6b65\u63d0\u51fa\u4e86\u591a\u4e13\u5bb6\u9009\u62e9\u673a\u5236\uff0c\u901a\u8fc7\u9009\u62e9\u5c11\u91cf\u4e13\u5bb6\u53c2\u4e0e\u8ba1\u7b97\uff0c\u964d\u4f4e\u4e86\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u3002\u7136\u800c\uff0c\u5373\u4f7f\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0cMoE\u4ecd\u7136\u9762\u4e34\u7740\u63a8\u7406\u5ef6\u8fdf\u8fc7\u9ad8\u7684\u95ee\u9898\uff0c\u7279\u522b\u662f\u5728\u9700\u8981\u5904\u7406\u5927\u6279\u91cf\u6570\u636e\u65f6\u3002<\/p>\n\n\n\n<p>\u76f8\u6bd4\u4e4b\u4e0b\uff0cPKM\uff08Lample et al., 2019\uff09\u91c7\u7528\u4e86\u4ea7\u54c1\u952e\uff08Product Keys\uff09\u673a\u5236\uff0c\u5c06\u5927\u89c4\u6a21\u7684\u5185\u5b58\u5c42\u5206\u5272\u4e3a\u591a\u4e2a\u5c0f\u5757\uff0c\u901a\u8fc7\u7d22\u5f15\u6765\u8bbf\u95ee\u7279\u5b9a\u7684\u503c\u3002\u8fd9\u79cd\u65b9\u6cd5\u80fd\u591f\u6709\u6548\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u7684\u9891\u7387\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u63a8\u7406\u6548\u7387\u3002\u7136\u800c\uff0cPKM\u7684\u6027\u80fd\u5e76\u672a\u8fbe\u5230MoE\u7684\u6c34\u5e73\uff0c\u5c24\u5176\u662f\u5728\u5927\u89c4\u6a21\u8bad\u7ec3\u7684\u573a\u666f\u4e0b\uff0cPKM\u4ecd\u7136\u9762\u4e34\u7740\u8bf8\u591a\u6311\u6218\u3002<\/p>\n\n\n\n<p>UltraMem\u7684\u63d0\u51fa\u6b63\u662f\u4e3a\u4e86\u514b\u670dMoE\u548cPKM\u7684\u5c40\u9650\u6027\uff0c\u91c7\u7528\u4e86\u6539\u8fdb\u7684\u67e5\u8be2-\u952e\u68c0\u7d22\u673a\u5236\uff0c\u4ee5\u53ca\u7a00\u758f\u6fc0\u6d3b\u548c\u5206\u5e03\u5f0f\u5185\u5b58\u8bbf\u95ee\u7b56\u7565\uff0c\u4f7f\u5f97\u6a21\u578b\u5728\u4fdd\u8bc1\u6027\u80fd\u7684\u540c\u65f6\uff0c\u80fd\u663e\u8457\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e09\u3001UltraMem\u67b6\u6784<\/h5>\n\n\n\n<p>UltraMem\u67b6\u6784\u5728PKM\u7684\u57fa\u7840\u4e0a\u8fdb\u884c\u4e86\u4e00\u7cfb\u5217\u6539\u8fdb\u3002\u9996\u5148\uff0cUltraMem\u5f15\u5165\u4e86Tucker\u5206\u89e3\uff08Tucker Decomposition\uff09\u6765\u4f18\u5316\u67e5\u8be2-\u952e\u68c0\u7d22\u8fc7\u7a0b\u3002Tucker\u5206\u89e3\u662f\u4e00\u79cd\u77e9\u9635\u5206\u89e3\u65b9\u6cd5\uff0c\u901a\u8fc7\u5c06\u5927\u89c4\u6a21\u7684\u67e5\u8be2-\u952e\u77e9\u9635\u5206\u89e3\u4e3a\u591a\u4e2a\u5c0f\u77e9\u9635\uff0c\u51cf\u5c11\u4e86\u8ba1\u7b97\u590d\u6742\u5ea6\u3002\u8fd9\u4f7f\u5f97UltraMem\u80fd\u591f\u5728\u5927\u89c4\u6a21\u5185\u5b58\u8bbf\u95ee\u7684\u60c5\u51b5\u4e0b\uff0c\u4ecd\u7136\u4fdd\u6301\u8f83\u4f4e\u7684\u8ba1\u7b97\u5f00\u9500\u3002<\/p>\n\n\n\n<p>\u5176\u6b21\uff0cUltraMem\u91c7\u7528\u4e86\u9690\u5f0f\u503c\u6269\u5c55\uff08Implicit Value Expansion\uff0cIVE\uff09\u6280\u672f\uff0c\u901a\u8fc7\u5c06\u5185\u5b58\u8868\u6269\u5c55\u4e3a\u591a\u4e2a\u865a\u62df\u5185\u5b58\u5757\uff0c\u8fdb\u4e00\u6b65\u51cf\u5c11\u4e86\u5185\u5b58\u8bbf\u95ee\u7684\u9700\u6c42\u3002\u5177\u4f53\u6765\u8bf4\uff0cIVE\u5c06\u6bcf\u4e2a\u5185\u5b58\u503c\u7684\u591a\u4e2a\u865a\u62df\u6269\u5c55\u6620\u5c04\u5230\u4e0d\u540c\u7684\u5185\u5b58\u5730\u5740\uff0c\u4ece\u800c\u5728\u63a8\u7406\u65f6\u53ef\u4ee5\u9009\u62e9\u6027\u5730\u8bbf\u95ee\u8fd9\u4e9b\u6269\u5c55\u503c\uff0c\u800c\u65e0\u9700\u76f4\u63a5\u8bbf\u95ee\u539f\u59cb\u5185\u5b58\u8868\u3002\u8fd9\u6837\uff0cUltraMem\u80fd\u591f\u5728\u4fdd\u6301\u8f83\u5c0f\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u7684\u540c\u65f6\uff0c\u5904\u7406\u66f4\u5927\u7684\u6a21\u578b\u548c\u6570\u636e\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0cUltraMem\u8fd8\u91c7\u7528\u4e86\u591a\u6838\u8bc4\u5206\uff08Multi-Core Scoring\uff0cMCS\uff09\u6280\u672f\uff0c\u5728\u5185\u5b58\u8bbf\u95ee\u65f6\u4e3a\u6bcf\u4e2a\u5185\u5b58\u503c\u5206\u914d\u591a\u4e2a\u8bc4\u5206\uff0c\u8fd9\u6709\u52a9\u4e8e\u63d0\u9ad8\u68c0\u7d22\u51c6\u786e\u6027\uff0c\u5e76\u8fdb\u4e00\u6b65\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u65f6\u7684\u5ef6\u8fdf\u3002\u901a\u8fc7\u8fd9\u4e9b\u6280\u672f\u7684\u7ec4\u5408\uff0cUltraMem\u80fd\u591f\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\u663e\u8457\u63d0\u9ad8\u901f\u5ea6\uff0c\u5e76\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u56db\u3001\u5b9a\u91cf\u5206\u6790<\/h5>\n\n\n\n<p>\u4e3a\u4f55\u9009\u62e9UltraMem\u800c\u975eMoE \u5728\u8bba\u6587\u4e2d\uff0c\u4f5c\u8005\u5bf9\u6bd4\u4e86UltraMem\u548cMoE\u7684\u6027\u80fd\uff0c\u7279\u522b\u662f\u5728\u63a8\u7406\u65f6\u7684\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u548c\u63a8\u7406\u901f\u5ea6\u3002\u5047\u8bbeTransformer\u6a21\u578b\u7684\u9690\u85cf\u5c42\u7ef4\u5ea6\u4e3aD\uff0cMoE\u4f7f\u7528\u6bcf\u4e2atoken\u9009\u62e9\u4e24\u4e2a\u4e13\u5bb6\uff0c\u5e76\u5047\u8bbe\u6bcf\u4e2a\u4e13\u5bb6\u7684\u5185\u90e8\u7ef4\u5ea6\u4e3a2D\u3002\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0cMoE\u7684\u5185\u5b58\u8bbf\u95ee\u91cf\u4e3a<code>min(2B, Nmoe) \u00d7 2D^2<\/code>\uff0c\u5176\u4e2dB\u662f\u6279\u6b21\u5927\u5c0f\uff0cNmoe\u662f\u4e13\u5bb6\u6570\u91cf\u3002\u800cUltraMem\u5219\u901a\u8fc7\u9009\u62e9\u524dm\u4e2a\u503c\u6765\u6fc0\u6d3b\u5185\u5b58\uff0c\u5185\u5b58\u8bbf\u95ee\u91cf\u4e3a<code>min(Bm, N) \u00d7 D\/2<\/code>\uff0c\u5176\u4e2dm\u662f\u6fc0\u6d3b\u7684\u503c\u7684\u6570\u91cf\u3002<\/p>\n\n\n\n<p>\u968f\u7740\u6279\u6b21\u5927\u5c0f\u7684\u589e\u52a0\uff0cMoE\u7684\u5185\u5b58\u8bbf\u95ee\u91cf\u8fc5\u901f\u589e\u957f\uff0c\u800cUltraMem\u7684\u5185\u5b58\u8bbf\u95ee\u91cf\u589e\u957f\u8f83\u6162\uff0c\u76f4\u5230\u8fbe\u5230\u4e0a\u9650\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u5e38\u89c1\u7684\u63a8\u7406\u6279\u6b21\u5927\u5c0f\u4e0b\uff0cUltraMem\u6bd4MoE\u5feb6\u500d\uff0c\u4e14\u5728\u5176\u4ed6\u6279\u6b21\u5927\u5c0f\u4e0b\u4e5f\u8868\u73b0\u51fa\u8272\u3002\u6b64\u5916\uff0c\u5728\u8f83\u5c0f\u6279\u6b21\u5927\u5c0f\u7684\u60c5\u51b5\u4e0b\uff0cMoE\u7684\u63a8\u7406\u901f\u5ea6\u663e\u8457\u4f4e\u4e8eUltraMem\uff0c\u8fdb\u4e00\u6b65\u8bc1\u660e\u4e86UltraMem\u5728\u5b9e\u9645\u63a8\u7406\u4e2d\u7684\u4f18\u52bf\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u4e94\u3001\u5b9e\u9a8c<\/h5>\n\n\n\n<p>\u5728\u5b9e\u9a8c\u90e8\u5206\uff0c\u4f5c\u8005\u5c55\u793a\u4e86UltraMem\u5728\u4e0d\u540c\u89c4\u6a21\u7684\u6a21\u578b\u4e2d\u7684\u8868\u73b0\uff0c\u7ed3\u679c\u8868\u660e\uff0cUltraMem\u80fd\u591f\u5728\u76f8\u540c\u7684\u8ba1\u7b97\u8d44\u6e90\u4e0b\uff0c\u8d85\u8d8aMoE\u548cPKM\uff0c\u5c24\u5176\u5728\u5904\u7406\u5927\u89c4\u6a21\u6a21\u578b\u65f6\u8868\u73b0\u5c24\u4e3a\u7a81\u51fa\u3002\u901a\u8fc7\u5bf9\u6bd4\u4e0d\u540c\u6a21\u578b\u7684\u63a8\u7406\u65f6\u95f4\u548c\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\uff0c\u4f5c\u8005\u9a8c\u8bc1\u4e86UltraMem\u7684\u4f18\u8d8a\u6027\u3002<\/p>\n\n\n\n<p>\u4f5c\u8005\u8fd8\u8fdb\u884c\u4e86\u5927\u91cf\u7684\u8bc4\u4f30\uff0c\u4f7f\u7528\u4e8610\u4e2a\u4e0d\u540c\u7684\u57fa\u51c6\u6570\u636e\u96c6\uff0c\u5305\u62ecMMLU\u3001TriviaQA\u3001GPQA\u7b49\uff0c\u8bc4\u4f30\u4e86\u6a21\u578b\u7684\u77e5\u8bc6\u80fd\u529b\u3001\u63a8\u7406\u80fd\u529b\u3001\u9605\u8bfb\u7406\u89e3\u80fd\u529b\u548c\u6574\u4f53\u8868\u73b0\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cUltraMem\u5728\u5404\u4e2a\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u5747\u4f18\u4e8eMoE\u548cPKM\uff0c\u5c24\u5176\u662f\u5728\u8ba1\u7b97\u548c\u5185\u5b58\u8d44\u6e90\u6709\u9650\u7684\u60c5\u51b5\u4e0b\uff0cUltraMem\u80fd\u591f\u63d0\u4f9b\u66f4\u9ad8\u7684\u63a8\u7406\u901f\u5ea6\u548c\u66f4\u4f18\u7684\u6a21\u578b\u6027\u80fd\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">\u516d\u3001\u7ed3\u8bba<\/h5>\n\n\n\n<p>\u8bba\u6587\u6700\u540e\u603b\u7ed3\u4e86UltraMem\u7684\u4f18\u52bf\u3002UltraMem\u901a\u8fc7\u51cf\u5c11\u5185\u5b58\u8bbf\u95ee\u5f00\u9500\u548c\u63d0\u9ad8\u63a8\u7406\u901f\u5ea6\uff0c\u663e\u8457\u63d0\u5347\u4e86\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\u7684\u63a8\u7406\u6027\u80fd\u3002\u4e0eMoE\u76f8\u6bd4\uff0cUltraMem\u4e0d\u4ec5\u5177\u6709\u76f8\u540c\u7684\u8ba1\u7b97\u8d44\u6e90\u6d88\u8017\uff0c\u8fd8\u5728\u63a8\u7406\u901f\u5ea6\u548c\u6027\u80fd\u4e0a\u5177\u6709\u660e\u663e\u7684\u4f18\u52bf\u3002\u901a\u8fc7\u5bf9UltraMem\u67b6\u6784\u7684\u6df1\u5165\u5206\u6790\uff0c\u8bba\u6587\u5c55\u793a\u4e86\u4e00\u79cd\u65b0\u7684\u9ad8\u6548\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b\u67b6\u6784\uff0c\u5b83\u4e3a\u672a\u6765\u5927\u89c4\u6a21\u6a21\u578b\u7684\u5f00\u53d1\u63d0\u4f9b\u4e86\u65b0\u7684\u601d\u8def\u548c\u65b9\u5411\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587Ultra-Sparse Memory Network\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3aUltraMem\u7684\u65b0\u578b\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c\u89e3 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,95,86],"class_list":["post-5660","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-transformer","tag-llm"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587Ultra-Sparse Memory Network\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3aUltraMem\u7684\u65b0\u578b\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c\u89e3&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5660","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5660"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5660\/revisions"}],"predecessor-version":[{"id":5662,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5660\/revisions\/5662"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5660"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5660"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5660"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}