{"id":5320,"date":"2023-10-26T15:35:00","date_gmt":"2023-10-26T07:35:00","guid":{"rendered":"https:\/\/ai-stack.ai\/?p=5320"},"modified":"2024-12-09T16:26:30","modified_gmt":"2024-12-09T08:26:30","slug":"ring-attention-processes-millions-words","status":"publish","type":"post","link":"https:\/\/ai-stack.ai\/zh-hant\/ring-attention-processes-millions-words","title":{"rendered":"\u53ef\u8655\u7406\u6578\u767e\u842c\u5b57\u8cc7\u6599\u91cf\u7684AI\u6a21\u578b: Ring Attention"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">\u5728\u4eba\u5de5\u667a\u80fd\uff08AI\uff09\u7684\u4e16\u754c\u88e1\uff0c\u6578\u64da\u662f\u4e00\u5207\u7684\u57fa\u790e\u3002\u4e0d\u904e\uff0c\u7576\u6211\u5011\u8ac7\u5230\u6578\u64da\u898f\u6a21\uff0c\u50b3\u7d71\u7684AI\u6a21\u578b\u5f80\u5f80\u6703\u9047\u5230\u74f6\u9838\u3002\u60f3\u50cf\u4e00\u4e0b\uff0c\u5982\u679c\u4f60\u6709\u4e00\u672c\u5305\u542b\u6578\u767e\u842c\u5b57\u7684\u66f8\uff0c\u800c\u4f60\u7684AI\u6a21\u578b\u53ea\u80fd\u8b80\u5e7e\u5343\u5b57\uff0c\u90a3\u6703\u662f\u591a\u9ebc\u7684\u5c40\u9650\uff01\u4f46\u73fe\u5728\uff0c\u6709\u4e86\u4e00\u7a2e\u540d\u70ba\u300cRing Attention\u300d\u7684\u65b0\u6280\u8853\uff0c\u9019\u4e00\u5207\u90fd\u6709\u4e86\u7ffb\u5929\u8986\u5730\u7684\u8b8a\u5316\u3002\u9019\u9805\u7531UC Berkeley\u548cGoogle DeepMind\u7684\u7814\u7a76\u4eba\u54e1\u958b\u767c\u7684\u6280\u8853\uff0c\u4e0d\u50c5\u89e3\u6c7a\u4e86\u8a18\u61b6\u9ad4\u9650\u5236\u7684\u554f\u984c\uff0c\u9084\u5927\u5927\u63d0\u9ad8\u4e86AI\u6a21\u578b\u5728\u8655\u7406\u5927\u898f\u6a21\u6578\u64da\u65b9\u9762\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h2-1\">\u50b3\u7d71\u8f49\u63db\u5668\uff08Transformers\uff09\u7684\u8a18\u61b6\u9ad4\u9650\u5236<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u8f49\u63db\u5668\uff08Transformers\uff09\u81ea\u5f9e\u51fa\u73fe\u4ee5\u4f86\uff0c\u5c31\u5728\u81ea\u7136\u8a9e\u8a00\u8655\u7406\uff08NLP\uff09\u548c\u6a5f\u5668\u5b78\u7fd2\uff08ML\uff09\u9818\u57df\u4e2d\u767c\u63ee\u4e86\u5de8\u5927\u7684\u4f5c\u7528\u3002\u7136\u800c\uff0c\u9019\u7a2e\u67b6\u69cb\u6709\u4e00\u500b\u660e\u986f\u7684\u7f3a\u9ede\uff1a\u5b83\u5011\u5728\u8655\u7406\u9577\u5e8f\u5217\u6578\u64da\u6642\u6703\u9047\u5230\u8a18\u61b6\u9ad4\u9650\u5236\u3002\u9019\u4e3b\u8981\u662f\u56e0\u70ba\u8f49\u63db\u5668\u4f7f\u7528\u4e86\u4e00\u7a2e\u540d\u70ba\u300c\u81ea\u6211\u6ce8\u610f\u529b\u300d\uff08self-attention\uff09\u7684\u6a5f\u5236\uff0c\u9019\u662f\u4e00\u500b\u975e\u5e38\u8a18\u61b6\u9ad4\u5bc6\u96c6\u578b\u7684\u904e\u7a0b\u3002\u50b3\u7d71\u4e0a\uff0c\u9019\u4e00\u9650\u5236\u4f7f\u5f97\u8f49\u63db\u5668\u96e3\u4ee5\u64f4\u5c55\u5176\u4e0a\u4e0b\u6587\u9577\u5ea6\uff0c\u5f9e\u800c\u9650\u5236\u4e86\u5b83\u5011\u5728\u8655\u7406\u5927\u898f\u6a21\u6578\u64da\u96c6\u65b9\u9762\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h2-3\">Ring Attention\uff1a\u4e00\u500b\u7a81\u7834\u6027\u7684\u89e3\u6c7a\u65b9\u6848<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u70ba\u4e86\u89e3\u6c7a\u9019\u500b\u554f\u984c\uff0cUC Berkeley\u7684\u7814\u7a76\u4eba\u54e1\u958b\u767c\u4e86\u4e00\u7a2e\u540d\u70ba\u300cRing Attention\u300d\u7684\u65b0\u65b9\u6cd5\u3002\u9019\u7a2e\u65b9\u6cd5\u7684\u6838\u5fc3\u601d\u60f3\u662f\u5c07\u8a08\u7b97\u904e\u7a0b\u5728\u591a\u500b\u8a2d\u5099\u4e4b\u9593\u4ee5\u584a\uff08block\uff09\u7684\u5f62\u5f0f\u5206\u4f48\u3002\u9019\u6a23\uff0c\u6bcf\u500b\u8a2d\u5099\u53ea\u9700\u8981\u8655\u7406\u4e00\u5c0f\u90e8\u5206\u7684\u6578\u64da\uff0c\u5f9e\u800c\u5927\u5927\u6e1b\u5c11\u4e86\u8a18\u61b6\u9ad4\u7684\u9700\u6c42\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u66f4\u5177\u9ad4\u5730\u8aaa\uff0cRing Attention\u63a1\u7528\u4e86\u4e00\u7a2e\u74b0\u72c0\uff08ring-like\uff09\u7684\u7d50\u69cb\uff0c\u5c07\u9375\u503c\uff08key-value\uff09\u584a\u5f9e\u4e00\u500b\u8a2d\u5099\u50b3\u9001\u5230\u53e6\u4e00\u500b\u8a2d\u5099\u3002\u9019\u7a2e\u584a\u5f0f\u7684\u6ce8\u610f\u529b\uff08blockwise attention\uff09\u548c\u524d\u5411\u904b\u7b97\uff08feedforward operations\uff09\u5141\u8a31\u6bcf\u500b\u8f38\u5165\u584a\u5177\u6709\u7279\u5b9a\u7684\u64cd\u4f5c\uff0c\u5f9e\u800c\u5be6\u73fe\u4e86\u9ad8\u6548\u7684\u8a08\u7b97\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h2-4\">\u5be6\u969b\u61c9\u7528\u548c\u672a\u4f86\u5c55\u671b<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u7a2e\u65b0\u65b9\u6cd5\u4e0d\u50c5\u514b\u670d\u4e86\u8a18\u61b6\u9ad4\u9650\u5236\uff0c\u9084\u4f7f\u5f97AI\u6a21\u578b\u80fd\u5920\u8655\u7406\u6bd4\u4ee5\u524d\u9577\u5f97\u591a\u7684\u5e8f\u5217\u3002\u6839\u64da\u7814\u7a76\u5831\u544a\uff0cRing Attention\u80fd\u5920\u8655\u7406\u7684\u5e8f\u5217\u9577\u5ea6\u6bd4\u5148\u524d\u7684\u8a18\u61b6\u9ad4\u9ad8\u6548\u6a21\u578b\u9577\u9054500\u500d\u3002\u9019\u610f\u5473\u8457\uff0c\u73fe\u5728\u7684AI\u6a21\u578b\u53ef\u4ee5\u8f15\u9b06\u8655\u7406\u6578\u767e\u842c\u5b57\u7684\u6578\u64da\u91cf\uff0c\u9019\u5c0d\u65bc\u5927\u898f\u6a21\u7684\u8996\u983b\u3001\u8a9e\u97f3\u548c\u8a9e\u8a00\u6a21\u578b\u4f86\u8aaa\u662f\u4e00\u500b\u5de8\u5927\u7684\u7a81\u7834\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u9805\u6280\u8853\u7684\u6f5b\u5728\u61c9\u7528\u7bc4\u570d\u975e\u5e38\u5ee3\u6cdb\uff0c\u5f9e\u5927\u898f\u6a21\u7684\u8996\u983b\u8a9e\u8a00\u6a21\u578b\u5230\u79d1\u5b78\u6578\u64da\uff0c\u5982\u57fa\u56e0\u5e8f\u5217\u7b49\u3002\u6b64\u5916\uff0c\u9019\u9805\u7814\u7a76\u9084\u70ba\u672a\u4f86\u63a2\u7d22\u6700\u5927\u5e8f\u5217\u9577\u5ea6\u548c\u6700\u5927\u8a08\u7b97\u6027\u80fd\u63d0\u4f9b\u4e86\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h2-5\">\u5982\u4f55\u5be6\u73feRing Attention\u7684\u6280\u8853<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5be6\u73feRing Attention\u7684\u95dc\u9375\u5728\u65bc\u5982\u4f55\u6709\u6548\u5730\u5c07\u8a08\u7b97\u904e\u7a0b\u5206\u4f48\u5230\u591a\u500b\u8a2d\u5099\u4e0a\u3002\u9019\u88e1\u6709\u5e7e\u500b\u5be6\u7528\u7684\u6b65\u9a5f\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u584a\u5283\u5206\uff08Block Partitioning\uff09\uff1a\u9996\u5148\uff0c\u5c07\u6574\u500b\u6578\u64da\u96c6\u5206\u6210\u591a\u500b\u5c0f\u584a\u3002<\/li>\n\n\n\n<li>\u74b0\u72c0\u7d50\u69cb\u8a2d\u8a08\uff08Ring Structure Design\uff09\uff1a\u78ba\u4fdd\u6240\u6709\u8a2d\u5099\u90fd\u6309\u7167\u4e00\u500b\u74b0\u72c0\u7d50\u69cb\u4f86\u6392\u5217\u3002<\/li>\n\n\n\n<li>\u9375\u503c\u584a\u50b3\u8f38\uff08Key-Value Block Transfer\uff09\uff1a\u5728\u9032\u884c\u8a08\u7b97\u7684\u540c\u6642\uff0c\u5c07\u9375\u503c\u584a\u5f9e\u4e00\u500b\u8a2d\u5099\u50b3\u9001\u5230\u53e6\u4e00\u500b\u8a2d\u5099\u3002<\/li>\n\n\n\n<li>\u584a\u5f0f\u6ce8\u610f\u529b\u548c\u524d\u5411\u904b\u7b97\uff08Blockwise Attention and Feedforward Operations\uff09\uff1a\u6bcf\u500b\u8a2d\u5099\u5c0d\u5176\u8ca0\u8cac\u7684\u6578\u64da\u584a\u9032\u884c\u6ce8\u610f\u529b\u548c\u524d\u5411\u904b\u7b97\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u6a23\uff0c\u6bcf\u500b\u8a2d\u5099\u53ea\u9700\u8981\u8ca0\u8cac\u4e00\u90e8\u5206\u7684\u8a08\u7b97\uff0c\u5f9e\u800c\u5927\u5927\u6e1b\u5c11\u4e86\u6574\u9ad4\u7684\u8a18\u61b6\u9ad4\u9700\u6c42\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"h2-6\">\u7d50\u8ad6\uff1a\u6253\u7834\u8a18\u61b6\u9ad4\u9650\u5236\uff0c\u958b\u555f\u7121\u7aae\u53ef\u80fd<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Ring Attention\u7684\u51fa\u73fe\u7121\u7591\u662fAI\u9818\u57df\u7684\u4e00\u500b\u91cd\u5927\u7a81\u7834\u3002\u5b83\u4e0d\u50c5\u89e3\u6c7a\u4e86\u9577\u671f\u56f0\u64fe\u7814\u7a76\u4eba\u54e1\u7684\u8a18\u61b6\u9ad4\u554f\u984c\uff0c\u9084\u70baAI\u6a21\u578b\u8655\u7406\u5927\u6578\u64da\u63d0\u4f9b\u4e86\u5168\u65b0\u7684\u53ef\u80fd\u3002\u5f9e\u73fe\u5728\u8d77\uff0c\u6578\u767e\u842c\u5b57\u7684\u6578\u64da\u91cf\u4e0d\u518d\u662f\u554f\u984c\uff0cAI\u7684\u61c9\u7528\u7bc4\u570d\u4e5f\u5c07\u56e0\u6b64\u800c\u5927\u5927\u64f4\u5c55\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7121\u8ad6\u662f\u5728\u5546\u696d\u61c9\u7528\u9084\u662f\u79d1\u5b78\u7814\u7a76\u4e2d\uff0cRing Attention\u90fd\u5c07\u6210\u70ba\u4e00\u500b\u4e0d\u53ef\u6216\u7f3a\u7684\u5de5\u5177\u3002\u5b83\u7684\u51fa\u73fe\u8b93\u6211\u5011\u770b\u5230\u4e86AI\u672a\u4f86\u7121\u7aae\u7684\u53ef\u80fd\u6027\uff0c\u8b93\u4eba\u5145\u6eff\u671f\u5f85\u3002\u6578\u64da\u96c6\u65b9\u9762\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>UC Berkeley\u548cGoogle DeepMind\u7684\u7814\u7a76\u5718\u968a\u958b\u767c\u4e86\u4e00\u7a2e\u540d\u70baRing Attention\u7684\u65b0\u6280\u8853\uff0c\u7a81\u7834\u4e86\u50b3\u7d71AI\u6a21\u578b\u5728\u8655\u7406\u5927\u898f\u6a21\u6578\u64da\u6642\u7684\u8a18\u61b6\u9ad4\u9650\u5236\u3002\u9019\u9805\u6280\u8853\u901a\u904e\u5c07\u8a08\u7b97\u904e\u7a0b\u5206\u4f48\u5230\u591a\u500b\u8a2d\u5099\uff0c\u6709\u6548\u63d0\u5347\u4e86AI\u6a21\u578b\u8655\u7406\u9577\u5e8f\u5217\u6578\u64da\u7684\u80fd\u529b\uff0c\u70baAI\u5728\u5404\u9818\u57df\u7684\u61c9\u7528\u958b\u95e2\u4e86\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/p>\n","protected":false},"author":253372381,"featured_media":5308,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_crdt_document":"","jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[96987581,96987573],"tags":[96987615],"class_list":["post-5320","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-focus","category-featured-articles","tag-ai"],"blocksy_meta":[],"acf":[],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/ai-stack.ai\/wp-content\/uploads\/2024\/03\/Featured-Article-05.jpg?fit=824%2C464&quality=100&ct=202603031250&ssl=1","jetpack_shortlink":"https:\/\/wp.me\/ph344V-1nO","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts\/5320","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/users\/253372381"}],"replies":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/comments?post=5320"}],"version-history":[{"count":0,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/posts\/5320\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/media\/5308"}],"wp:attachment":[{"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/media?parent=5320"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/categories?post=5320"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ai-stack.ai\/zh-hant\/wp-json\/wp\/v2\/tags?post=5320"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}