{"id":5657,"date":"2025-02-19T17:15:24","date_gmt":"2025-02-19T09:15:24","guid":{"rendered":"https:\/\/nullthought.net\/?p=5657"},"modified":"2025-02-23T11:08:03","modified_gmt":"2025-02-23T03:08:03","slug":"token-statistics-transformer%ef%bc%9a%e9%80%9a%e8%bf%87%e5%bc%95%e5%85%a5tssa%ef%bc%88token-statistics-self-attention%ef%bc%89%e6%b3%a8%e6%84%8f%e5%8a%9b%e6%a8%a1%e5%9d%97%ef%bc%8c%e6%98%be%e8%91%97","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=5657","title":{"rendered":"Token Statistics Transformer\uff1a\u901a\u8fc7\u5f15\u5165TSSA\uff08Token Statistics Self-Attention\uff09\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u663e\u8457\u964d\u4f4eTransformer\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\u548c\u5185\u5b58\u9700\u6c42"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2412.17810\" target=\"_blank\" rel=\"noreferrer noopener\">Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction<\/a><\/strong>\u63d0\u51fa\u7684Token Statistics Transformer\uff08ToST\uff09\u901a\u8fc7\u5f15\u5165TSSA\uff08Token Statistics Self-Attention\uff09\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u663e\u8457\u964d\u4f4e\u4e86\u8ba1\u7b97\u590d\u6742\u5ea6\u548c\u5185\u5b58\u9700\u6c42\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u4e0e\u4f20\u7edfTransformer\u67b6\u6784\u76f8\u5f53\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8bba\u6587\u4f5c\u8005\u4e3aZiyang Wu, Tianjiao Ding, Yifu Lu, Druv Pai, Jingyuan Zhang, Weida Wang, Yaodong Yu, Yi Ma, Benjamin D. Haeffele\uff0c\u6765\u81eaUC Berkeley, UPenn, UMich, THU, HKU, JHU\u7b49\u673a\u6784\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"981\" height=\"489\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-11.png\" alt=\"\" class=\"wp-image-5658\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-11.png 981w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-11-300x150.png 300w, https:\/\/nullthought.net\/wp-content\/uploads\/2025\/02\/image-11-768x383.png 768w\" sizes=\"auto, (max-width: 981px) 100vw, 981px\" \/><figcaption class=\"wp-element-caption\"><strong><a href=\"https:\/\/arxiv.org\/abs\/2412.17810\" target=\"_blank\" rel=\"noreferrer noopener\">Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u4e00\u3001\u7814\u7a76\u52a8\u673a\u4e0e\u80cc\u666f<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Transformer\u67b6\u6784\u81ea2017\u5e74\u63d0\u51fa\u4ee5\u6765\uff0c\u5df2\u7ecf\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u8ba1\u7b97\u673a\u89c6\u89c9\u7b49\u591a\u4e2a\u9886\u57df\u53d6\u5f97\u4e86\u6781\u5927\u7684\u6210\u529f\u3002\u5176\u6838\u5fc3\u521b\u65b0\u4e4b\u4e00\u4fbf\u662f\u81ea\u6ce8\u610f\u529b\u673a\u5236\uff08Self-Attention\uff09\uff0c\u8fd9\u4f7f\u5f97Transformer\u80fd\u591f\u9ad8\u6548\u5730\u5904\u7406\u8f93\u5165\u5e8f\u5217\u4e2d\u7684\u957f\u8ddd\u79bb\u4f9d\u8d56\u3002\u7136\u800c\uff0c\u5c3d\u7ba1\u81ea\u6ce8\u610f\u529b\u673a\u5236\u5728\u591a\u4e2a\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u4f46\u5176\u8ba1\u7b97\u590d\u6742\u5ea6\u662f\u4e00\u4e2a\u663e\u8457\u7684\u74f6\u9888\u3002\u4f20\u7edf\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236\u9700\u8981\u8ba1\u7b97\u6240\u6709\u6807\u8bb0\u5bf9\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\uff0c\u56e0\u6b64\u5176\u8ba1\u7b97\u590d\u6742\u5ea6\u662f\u5e73\u65b9\u7ea7\u522b\u7684\u3002\u968f\u7740\u8f93\u5165\u6807\u8bb0\u6570\u91cf\u7684\u589e\u52a0\uff0c\u8ba1\u7b97\u8d44\u6e90\u9700\u6c42\u4e5f\u968f\u4e4b\u5927\u5e45\u5ea6\u63d0\u5347\uff0c\u5c24\u5176\u662f\u5728\u5904\u7406\u957f\u5e8f\u5217\u4efb\u52a1\u65f6\uff0c\u8fd9\u4e00\u95ee\u9898\u5c24\u4e3a\u7a81\u51fa\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8be5\u8bba\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u6ce8\u610f\u529b\u673a\u5236\u2014\u2014<strong>Token Statistics Self-Attention (TSSA)<\/strong>\uff0c\u901a\u8fc7\u5f15\u5165\u53d8\u5206\u6700\u5927\u7f16\u7801\u7387\u964d\u7ef4\uff08MCR2\uff09\u76ee\u6807\uff0c\u8bbe\u8ba1\u4e86\u4e00\u79cd\u7ebf\u6027\u65f6\u95f4\u590d\u6742\u5ea6\u7684\u6ce8\u610f\u529b\u64cd\u4f5c\u7b26\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u6cd5\uff0cTransformer\u67b6\u6784\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\u53ef\u4ee5\u4ece\u4f20\u7edf\u7684\u4e8c\u6b21\u65b9\u590d\u6742\u5ea6\u964d\u4f4e\u5230\u7ebf\u6027\u590d\u6742\u5ea6\uff0c\u4ece\u800c\u5927\u5927\u63d0\u9ad8\u4e86\u8ba1\u7b97\u6548\u7387\uff0c\u540c\u65f6\u4fdd\u6301\u4e86\u4e0e\u4f20\u7edfTransformer\u67b6\u6784\u76f8\u5f53\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u4e8c\u3001\u65b9\u6cd5\u8bba\u4e0e\u65b0\u578b\u6ce8\u610f\u529b\u673a\u5236<\/strong><\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u6700\u5927\u7f16\u7801\u7387\u964d\u7ef4\uff08MCR2\uff09\u76ee\u6807\u4e0e\u53d8\u5206\u5f62\u5f0f<\/strong>\u5728Transformer\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4e2d\uff0c\u6838\u5fc3\u64cd\u4f5c\u662f\u8ba1\u7b97\u8f93\u5165\u6807\u8bb0\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\uff0c\u5e76\u57fa\u4e8e\u8fd9\u4e9b\u76f8\u4f3c\u5ea6\u52a0\u6743\u5e73\u5747\u751f\u6210\u8f93\u51fa\u6807\u8bb0\u3002\u8fd9\u4e00\u64cd\u4f5c\u9700\u8981\u8ba1\u7b97\u6bcf\u4e00\u5bf9\u6807\u8bb0\u7684\u76f8\u4f3c\u5ea6\uff0c\u5bfc\u81f4\u8ba1\u7b97\u590d\u6742\u5ea6\u4e3aO(n\u00b2)\uff0c\u5176\u4e2dn\u662f\u8f93\u5165\u6807\u8bb0\u7684\u6570\u91cf\u3002\u4e3a\u4e86\u51cf\u5c11\u8ba1\u7b97\u590d\u6742\u5ea6\uff0c\u8bba\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u578b\u7684\u4f18\u5316\u76ee\u6807\u2014\u2014<strong>\u6700\u5927\u7f16\u7801\u7387\u964d\u7ef4\uff08MCR2\uff09<\/strong>\u76ee\u6807\u3002\u8be5\u76ee\u6807\u901a\u8fc7\u538b\u7f29\u6807\u8bb0\u7279\u5f81\u540c\u65f6\u6269\u5c55\u6574\u4f53\u7279\u5f81\u7684\u7a7a\u95f4\uff0c\u4ece\u800c\u6709\u6548\u5730\u51cf\u5c11\u8ba1\u7b97\u590d\u6742\u5ea6\u3002MCR2\u76ee\u6807\u4e2d\u7684\u538b\u7f29\u9879\u65e8\u5728\u5c06\u540c\u4e00\u7ec4\u6807\u8bb0\u7684\u7279\u5f81\u538b\u7f29\u5230\u4e00\u4e2a\u4f4e\u7ef4\u7a7a\u95f4\u4e2d\uff0c\u800c\u6269\u5c55\u9879\u5219\u8bd5\u56fe\u5c06\u6240\u6709\u6807\u8bb0\u7684\u7279\u5f81\u5c3d\u53ef\u80fd\u5730\u6269\u5c55\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u7ed9\u5b9a\u4e00\u4e2a\u6807\u8bb0\u77e9\u9635X\uff0c\u8bba\u6587\u5047\u8bbe\u8fd9\u4e9b\u6807\u8bb0\u5c5e\u4e8eK\u4e2a\u7ec4\uff08\u4f8b\u5982\uff0c\u8868\u793a\u4e0d\u540c\u7684\u6807\u8bb0\u6a21\u5f0f\uff09\uff0cMCR2\u76ee\u6807\u65e8\u5728\u627e\u5230\u6bcf\u4e2a\u7ec4\u5185\u7279\u5f81\u7684\u538b\u7f29\u8868\u793a\uff0c\u540c\u65f6\u4fdd\u6301\u6574\u4f53\u6807\u8bb0\u7279\u5f81\u7684\u6269\u5c55\u6027\u3002\u901a\u8fc7\u6700\u5927\u5316MCR2\u76ee\u6807\uff0c\u6a21\u578b\u53ef\u4ee5\u81ea\u52a8\u5b66\u4e60\u5230\u6570\u636e\u7684\u4f4e\u7ef4\u7ed3\u6784\u3002\u8bba\u6587\u8fdb\u4e00\u6b65\u63a8\u5bfc\u4e86MCR2\u76ee\u6807\u7684\u53d8\u5206\u5f62\u5f0f\uff0c\u5e76\u4f7f\u7528\u201c\u767d\u76d2\u201d\u67b6\u6784\u8bbe\u8ba1\u65b9\u6cd5\u5c06\u5176\u8f6c\u5316\u4e3a\u4e00\u79cd\u65b0\u7684\u6ce8\u610f\u529b\u673a\u5236\u3002\u8fd9\u4e00\u6ce8\u610f\u529b\u673a\u5236\u4e0d\u518d\u4f9d\u8d56\u4e8e\u6807\u8bb0\u5bf9\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u8ba1\u7b97\uff0c\u800c\u662f\u57fa\u4e8e\u6807\u8bb0\u7279\u5f81\u7684\u7edf\u8ba1\u91cf\uff08\u5982\u4e8c\u9636\u77e9\uff09\u6765\u8fdb\u884c\u8ba1\u7b97\uff0c\u4ece\u800c\u5927\u5927\u63d0\u9ad8\u4e86\u8ba1\u7b97\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u767d\u76d2\u7f51\u7edc\u67b6\u6784\u8bbe\u8ba1<\/strong>\u201c\u767d\u76d2\u201d\u67b6\u6784\u8bbe\u8ba1\u662f\u4e00\u79cd\u901a\u8fc7\u9010\u6b65\u4f18\u5316\u76ee\u6807\u51fd\u6570\u6765\u6784\u5efa\u7f51\u7edc\u7684\u65b9\u5f0f\u3002\u5728\u8fd9\u79cd\u8bbe\u8ba1\u65b9\u6cd5\u4e2d\uff0c\u7f51\u7edc\u7684\u6bcf\u4e00\u5c42\u90fd\u53ef\u4ee5\u770b\u4f5c\u662f\u6267\u884c\u4e00\u4e2a\u4f18\u5316\u6b65\u9aa4\uff0c\u8be5\u6b65\u9aa4\u65e8\u5728\u6700\u5c0f\u5316\u6216\u6700\u5927\u5316\u67d0\u4e2a\u76ee\u6807\u51fd\u6570\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u7f51\u7edc\u67b6\u6784\u7684\u8bbe\u8ba1\u8fc7\u7a0b\u53ef\u4ee5\u4e0e\u4f18\u5316\u7b97\u6cd5\uff08\u5982\u68af\u5ea6\u4e0b\u964d\uff09\u76f8\u8054\u7cfb\uff0c\u4f7f\u5f97\u6bcf\u4e00\u5c42\u7684\u64cd\u4f5c\u90fd\u53ef\u4ee5\u901a\u8fc7\u4f18\u5316\u6b65\u9aa4\u6765\u89e3\u91ca\u3002\u8bba\u6587\u5728\u767d\u76d2\u67b6\u6784\u8bbe\u8ba1\u7684\u57fa\u7840\u4e0a\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u6ce8\u610f\u529b\u64cd\u4f5c\u7b26\uff0c\u79f0\u4e3a<strong>Token Statistics Self-Attention (TSSA)<\/strong>\u3002\u4e0e\u4f20\u7edf\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4e0d\u540c\uff0cTSSA\u4e0d\u518d\u8ba1\u7b97\u6bcf\u5bf9\u6807\u8bb0\u7684\u76f8\u4f3c\u5ea6\uff0c\u800c\u662f\u901a\u8fc7\u6807\u8bb0\u7279\u5f81\u7684\u4e8c\u9636\u77e9\u7edf\u8ba1\u91cf\u6765\u8fdb\u884c\u4f4e\u79e9\u6295\u5f71\u3002\u8fd9\u4e00\u65b9\u6cd5\u901a\u8fc7\u5bf9\u8f93\u5165\u6807\u8bb0\u7279\u5f81\u8fdb\u884c\u6570\u636e\u4f9d\u8d56\u7684\u4f4e\u79e9\u6295\u5f71\uff0c\u53ea\u4fdd\u7559\u91cd\u8981\u4fe1\u606f\uff0c\u6291\u5236\u4e0d\u91cd\u8981\u7684\u4fe1\u606f\uff0c\u4ece\u800c\u6709\u6548\u5730\u51cf\u5c11\u4e86\u8ba1\u7b97\u8d1f\u62c5\u3002\u8bba\u6587\u4e2d\u8fd8\u6307\u51fa\uff0cTSSA\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\u4e3aO(n)\uff0c\u5373\u4e0e\u6807\u8bb0\u7684\u6570\u91cf\u5448\u7ebf\u6027\u5173\u7cfb\uff0c\u8fd9\u4f7f\u5f97TSSA\u6210\u4e3a\u4e00\u4e2a\u975e\u5e38\u9ad8\u6548\u7684\u6ce8\u610f\u529b\u64cd\u4f5c\u7b26\u3002<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u4e09\u3001\u5b9e\u9a8c\u7ed3\u679c\u4e0e\u5206\u6790<\/strong><\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u89c6\u89c9\u4efb\u52a1<\/strong>\u4e3a\u4e86\u9a8c\u8bc1\u6240\u63d0\u51fa\u7684Token Statistics Transformer (TOST)\u67b6\u6784\uff0c\u8bba\u6587\u5bf9\u5176\u5728\u89c6\u89c9\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u8fdb\u884c\u4e86\u8bc4\u4f30\u3002\u5b9e\u9a8c\u4e2d\u4f7f\u7528\u4e86\u591a\u4e2a\u6807\u51c6\u6570\u636e\u96c6\uff0c\u5982ImageNet\u3001CIFAR10\u3001Oxford Flowers\u7b49\uff0c\u5e76\u5c06TOST\u4e0e\u4f20\u7edf\u7684Vision Transformer\uff08ViT\uff09\u67b6\u6784\u8fdb\u884c\u4e86\u6bd4\u8f83\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cTOST\u5728\u8fd9\u4e9b\u6570\u636e\u96c6\u4e0a\u7684\u5206\u7c7b\u51c6\u786e\u7387\u4e0eViT\u76f8\u5f53\uff0c\u4f46\u5176\u8ba1\u7b97\u548c\u5185\u5b58\u590d\u6742\u5ea6\u663e\u8457\u4f4e\u4e8eViT\u3002\u7279\u522b\u5730\uff0c\u5728\u5904\u7406\u5927\u89c4\u6a21\u56fe\u50cf\u6570\u636e\u96c6\u65f6\uff0cTOST\u7684\u6548\u7387\u4f18\u52bf\u66f4\u52a0\u660e\u663e\u3002\u901a\u8fc7\u66ff\u6362\u6807\u51c6\u7684\u81ea\u6ce8\u610f\u529b\u6a21\u5757\uff0cTOST\u80fd\u591f\u5728\u4fdd\u6301\u76f8\u4f3c\u6027\u80fd\u7684\u540c\u65f6\uff0c\u663e\u8457\u51cf\u5c11\u8ba1\u7b97\u548c\u5185\u5b58\u6d88\u8017\u3002<\/li>\n\n\n\n<li><strong>\u957f\u5e8f\u5217\u5efa\u6a21<\/strong>\u8bba\u6587\u8fd8\u5728\u957f\u5e8f\u5217\u5efa\u6a21\u4efb\u52a1\u4e0a\u5bf9TOST\u8fdb\u884c\u4e86\u8bc4\u4f30\uff0c\u4f7f\u7528\u4e86<strong>Long-Range Arena (LRA)<\/strong>\u57fa\u51c6\u3002LRA\u57fa\u51c6\u65e8\u5728\u6d4b\u8bd5\u6a21\u578b\u5728\u5904\u7406\u957f\u5e8f\u5217\u65f6\u7684\u80fd\u529b\uff0c\u6db5\u76d6\u4e86\u591a\u4e2a\u957f\u6587\u6863\u7406\u89e3\u4efb\u52a1\uff0c\u5982\u6587\u672c\u68c0\u7d22\u3001\u56fe\u50cf\u8def\u5f84\u67e5\u627e\u7b49\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cTOST\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u5c24\u5176\u5728\u957f\u5e8f\u5217\u5efa\u6a21\u65b9\u9762\uff0c\u5176\u6027\u80fd\u8d85\u8fc7\u4e86\u5927\u591a\u6570\u73b0\u6709\u7684Transformer\u53d8\u79cd\u3002\u4e0e\u4f20\u7edfTransformer\u6a21\u578b\u76f8\u6bd4\uff0cTOST\u5728\u957f\u5e8f\u5217\u4efb\u52a1\u4e2d\u5c55\u73b0\u4e86\u66f4\u9ad8\u7684\u6548\u7387\u548c\u66f4\u4f4e\u7684\u8ba1\u7b97\u9700\u6c42\u3002<\/li>\n\n\n\n<li><strong>\u8bed\u8a00\u5efa\u6a21\u4efb\u52a1<\/strong>\u8bba\u6587\u8fd8\u8bc4\u4f30\u4e86TOST\u5728\u8bed\u8a00\u5efa\u6a21\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u3002\u901a\u8fc7\u5c06TOST\u5e94\u7528\u4e8e\u6807\u51c6\u7684\u8bed\u8a00\u5efa\u6a21\u6570\u636e\u96c6\uff08\u5982OpenWebText\u3001WikiText\u7b49\uff09\uff0c\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cTOST\u80fd\u591f\u5728\u4e0d\u727a\u7272\u6027\u80fd\u7684\u524d\u63d0\u4e0b\uff0c\u663e\u8457\u63d0\u9ad8\u8ba1\u7b97\u6548\u7387\u3002\u5c3d\u7ba1\u81ea\u6ce8\u610f\u529b\u673a\u5236\u901a\u5e38\u88ab\u8ba4\u4e3a\u662f\u8bed\u8a00\u5efa\u6a21\u6210\u529f\u7684\u5173\u952e\uff0cTOST\u901a\u8fc7\u66ff\u6362\u81ea\u6ce8\u610f\u529b\u64cd\u4f5c\u7b26\uff0c\u4ecd\u80fd\u4fdd\u6301\u8f83\u597d\u7684\u6027\u80fd\uff0c\u5e76\u5728\u8ba1\u7b97\u901f\u5ea6\u548c\u5185\u5b58\u4f7f\u7528\u65b9\u9762\u5177\u6709\u660e\u663e\u4f18\u52bf\u3002<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u56db\u3001\u603b\u7ed3\u4e0e\u5c55\u671b<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u4e2a\u521b\u65b0\u7684Token Statistics Transformer\u67b6\u6784\uff0c\u901a\u8fc7\u5f15\u5165Token Statistics Self-Attention (TSSA)\u6a21\u5757\uff0c\u89e3\u51b3\u4e86\u6807\u51c6Transformer\u67b6\u6784\u4e2d\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97\u590d\u6742\u5ea6\u8fc7\u9ad8\u7684\u95ee\u9898\u3002TSSA\u901a\u8fc7\u5229\u7528\u6570\u636e\u4f9d\u8d56\u7684\u4f4e\u79e9\u6295\u5f71\uff0c\u663e\u8457\u964d\u4f4e\u4e86\u8ba1\u7b97\u590d\u6742\u5ea6\uff0c\u4e14\u4e0d\u4f9d\u8d56\u4e8e\u6807\u8bb0\u5bf9\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u8ba1\u7b97\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cTOST\u5728\u591a\u4e2a\u89c6\u89c9\u3001\u8bed\u8a00\u548c\u957f\u5e8f\u5217\u4efb\u52a1\u4e2d\u90fd\u8868\u73b0\u51fa\u8272\uff0c\u5e76\u4e14\u5728\u8ba1\u7b97\u6548\u7387\u548c\u5185\u5b58\u4f7f\u7528\u65b9\u9762\u8fdc\u8fdc\u4f18\u4e8e\u4f20\u7edf\u7684Transformer\u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u672a\u6765\u7684\u5de5\u4f5c\u53ef\u4ee5\u8fdb\u4e00\u6b65\u4f18\u5316TSSA\u6a21\u5757\u7684\u8ba1\u7b97\u6548\u7387\uff0c\u63a2\u7d22\u5176\u5728\u66f4\u591a\u4efb\u52a1\u4e2d\u7684\u5e94\u7528\uff0c\u5e76\u7ed3\u5408\u66f4\u9ad8\u6548\u7684\u8bad\u7ec3\u65b9\u6cd5\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u5176\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u8868\u73b0\u3002TOSS\u67b6\u6784\u7684\u63d0\u51fa\u4e3aTransformer\u6a21\u578b\u7684\u9ad8\u6548\u8bbe\u8ba1\u63d0\u4f9b\u4e86\u65b0\u7684\u601d\u8def\uff0c\u5e76\u4e14\u6709\u6f5c\u529b\u5728\u591a\u4e2a\u5b9e\u9645\u4efb\u52a1\u4e2d\u5f97\u5230\u5e7f\u6cdb\u5e94\u7528\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"wp-block-paragraph\">Token Statistics Transformer\uff08ToST\uff09: <a href=\"https:\/\/robinwu218.github.io\/ToST\/\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/robinwu218.github.io\/ToST\/<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ToST on GitHub: <a href=\"https:\/\/github.com\/RobinWu218\/ToST\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/RobinWu218\/ToST<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587Token Statistics Transformer: Linear-Time Attention v [&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":[8,36],"tags":[39,95],"class_list":["post-5657","post","type-post","status-publish","format-standard","hentry","category-tech","category-36","tag-ai","tag-transformer"],"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> <a href=\"https:\/\/nullthought.net\/?cat=36\" rel=\"category\">\u79d1\u5b66<\/a>","rttpg_excerpt":"\u8bba\u6587Token Statistics Transformer: Linear-Time Attention v&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5657","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=5657"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5657\/revisions"}],"predecessor-version":[{"id":5659,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/5657\/revisions\/5659"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5657"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5657"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5657"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}