Notice: Function _load_textdomain_just_in_time was called incorrectly. Translation loading for the betterdocs domain was triggered too early. This is usually an indicator for some code in the plugin or theme running too early. Translations should be loaded at the init action or later. Please see Debugging in WordPress for more information. (This message was added in version 6.7.0.) in /data/user/htdocs/wp-includes/functions.php on line 6114

Notice: Function _load_textdomain_just_in_time was called incorrectly. Translation loading for the jnews-view-counter domain was triggered too early. This is usually an indicator for some code in the plugin or theme running too early. Translations should be loaded at the init action or later. Please see Debugging in WordPress for more information. (This message was added in version 6.7.0.) in /data/user/htdocs/wp-includes/functions.php on line 6114

Notice: Function _load_textdomain_just_in_time was called incorrectly. Translation loading for the wp-statistics domain was triggered too early. This is usually an indicator for some code in the plugin or theme running too early. Translations should be loaded at the init action or later. Please see Debugging in WordPress for more information. (This message was added in version 6.7.0.) in /data/user/htdocs/wp-includes/functions.php on line 6114

Notice: Function _load_textdomain_just_in_time was called incorrectly. Translation loading for the wpdiscuz domain was triggered too early. This is usually an indicator for some code in the plugin or theme running too early. Translations should be loaded at the init action or later. Please see Debugging in WordPress for more information. (This message was added in version 6.7.0.) in /data/user/htdocs/wp-includes/functions.php on line 6114

Notice: 函数 _load_textdomain_just_in_time 的调用方法不正确jnews 域的翻译加载触发过早。这通常表示插件或主题中的某些代码运行过早。翻译应在 init 操作或之后加载。 请查阅调试 WordPress来获取更多信息。 (这个消息是在 6.7.0 版本添加的。) in /data/user/htdocs/wp-includes/functions.php on line 6114

Notice: 函数 _load_textdomain_just_in_time 的调用方法不正确jnews-like 域的翻译加载触发过早。这通常表示插件或主题中的某些代码运行过早。翻译应在 init 操作或之后加载。 请查阅调试 WordPress来获取更多信息。 (这个消息是在 6.7.0 版本添加的。) in /data/user/htdocs/wp-includes/functions.php on line 6114

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893

Warning: Cannot modify header information - headers already sent by (output started at /data/user/htdocs/wp-includes/functions.php:6114) in /data/user/htdocs/wp-includes/rest-api/class-wp-rest-server.php on line 1893
{"id":34144,"date":"2024-09-22T01:44:15","date_gmt":"2024-09-21T17:44:15","guid":{"rendered":"https:\/\/linguaresources.com\/?p=34144"},"modified":"2024-09-22T01:44:15","modified_gmt":"2024-09-21T17:44:15","slug":"%e5%a4%a7%e6%a8%a1%e5%9e%8b%e7%8e%a9%e3%80%8a%e9%bb%91%e7%a5%9e%e8%af%9d%ef%bc%9a%e6%82%9f%e7%a9%ba%e3%80%8b%ef%bc%8c%e5%ae%8c%e6%88%90-90-%e7%ae%80%e5%8d%95%e3%80%81%e4%b8%ad%e7%ad%89%e6%b0%b4","status":"publish","type":"post","link":"https:\/\/linguaresources.com\/?p=34144","title":{"rendered":"\u5927\u6a21\u578b\u73a9\u300a\u9ed1\u795e\u8bdd\uff1a\u609f\u7a7a\u300b\uff0c\u5b8c\u6210 90% \u7b80\u5355\u3001\u4e2d\u7b49\u6c34\u5e73\u6218\u6597"},"content":{"rendered":"
\n

\u5927\u6a21\u578b\u73a9\u300a\u9ed1\u795e\u8bdd\uff1a\u609f\u7a7a\u300b\uff0c\u5b8c\u6210 90% \u7b80\u5355\u3001\u4e2d\u7b49\u6c34\u5e73\u6218\u6597<\/strong><\/h1>\n
<\/a>\u7ffb\u8bd1\u6280\u672f\u6559\u80b2\u4e0e\u7814\u7a76<\/span><\/span><\/p>\n
<\/section>\n

 <\/p>\n

2024\u5e7409\u670822\u65e5 00:02<\/em>\u00a0<\/span>\u9655\u897f<\/span><\/em><\/span><\/section>\n

\n

\"\"<\/p>\n

\u6700\u8fd1\uff0c\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u667a\u80fd\u4f53\u5728\u5404\u4e2a\u9886\u57df\u90fd\u53d6\u5f97\u4e86\u91cd\u5927\u8fdb\u5c55\u3002\u6700\u70ed\u95e8\u7684\u7814\u7a76\u9886\u57df\u4e4b\u4e00\u662f\u5c06\u8fd9\u4e9b\u667a\u80fd\u4f53\u5e94\u7528\u4e8e\u89c6\u9891\u6e38\u620f\u4e2d\u3002\u8fd9\u4e9b\u65b9\u6cd5\u901a\u5e38\u4f9d\u8d56\u4e8e\u6e38\u620f API \u6765\u8bbf\u95ee\u6e38\u620f\u4e2d\u7684\u73af\u5883\u548c\u52a8\u4f5c\u6570\u636e\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u53d7\u9650\u4e8e API \u7684\u53ef\u7528\u6027\uff0c\u65e0\u6cd5\u53cd\u6620\u4eba\u7c7b\u73a9\u6e38\u620f\u7684\u65b9\u5f0f\u3002<\/span><\/section>\n
\u968f\u7740\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u7684\u51fa\u73b0\uff0c\u667a\u80fd\u4f53\u73b0\u5728\u62e5\u6709\u4e86\u66f4\u5f3a\u7684\u89c6\u89c9\u7406\u89e3\u80fd\u529b\uff0c\u4f7f\u5176\u80fd\u591f\u4ec5\u4f7f\u7528\u89c6\u89c9\u8f93\u5165\u4e0e\u6e38\u620f\u8fdb\u884c\u4ea4\u4e92\u3002\u5c3d\u7ba1\u53d6\u5f97\u4e86\u8fd9\u4e9b\u8fdb\u6b65\uff0c\u4f46\u76ee\u524d\u7684\u65b9\u6cd5\u5728\u9762\u5411\u52a8\u4f5c\u7684\u4efb\u52a1\u4e2d\u4ecd\u9762\u4e34\u6311\u6218\uff0c\u7279\u522b\u662f\u5728\u52a8\u4f5c\u89d2\u8272\u626e\u6f14\u6e38\u620f\uff08ARPG\uff09\u4e2d\uff0c\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5\u975e\u5e38\u666e\u904d\uff0c\u4f46\u901a\u7528\u6027\u5dee\uff0c\u9700\u8981\u5927\u91cf\u8bad\u7ec3\u3002<\/span><\/section>\n
\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u963f\u91cc\u56e2\u961f\u9009\u62e9\u4ee5\u300a\u9ed1\u795e\u8bdd\uff1a\u609f\u7a7a\u300b\u4f5c\u4e3a\u7814\u7a76\u5e73\u53f0\uff0c\u63a2\u7d22\u73b0\u6709 VLM \u5728\u9700\u8981\u7eaf\u89c6\u89c9\u8f93\u5165\u548c\u590d\u6742\u52a8\u4f5c\u8f93\u51fa\u7684\u573a\u666f\u4e2d\u7684\u80fd\u529b\u8fb9\u754c\u3002\u4ed6\u4eec\u5728\u6e38\u620f\u4e2d\u5b9a\u4e49\u4e86 12 \u9879\u4efb\u52a1\uff0c\u5176\u4e2d 75% \u4ee5\u6218\u6597\u4e3a\u4e3b\uff0c\u5e76\u5c06\u51e0\u79cd SOTA \u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7eb3\u5165\u8fd9\u4e00\u57fa\u51c6\u3002\u6b64\u5916\uff0c\u4ed6\u4eec\u8fd8\u5c06\u53d1\u5e03\u4e00\u4e2a\u4eba\u5de5\u64cd\u4f5c\u6570\u636e\u96c6\uff0c\u5176\u4e2d\u5305\u542b\u5f55\u5236\u7684\u6e38\u620f\u89c6\u9891\u548c\u64cd\u4f5c\u65e5\u5fd7\uff0c\u5305\u62ec\u9f20\u6807\u548c\u952e\u76d8\u64cd\u4f5c\u3002\u4ed6\u4eec\u8fd8\u63d0\u51fa\u4e86\u4e00\u4e2a VARP\uff08\u89c6\u89c9\u52a8\u4f5c\u89d2\u8272\u626e\u6f14\uff09\u667a\u80fd\u4f53\u6846\u67b6\uff0c\u7531\u52a8\u4f5c\u89c4\u5212\u7cfb\u7edf\u548c\u89c6\u89c9\u8f68\u8ff9\u7cfb\u7edf\u7ec4\u6210\u3002\u8fd9\u4e00\u6846\u67b6\u5c55\u793a\u4e86\u6267\u884c\u57fa\u672c\u4efb\u52a1\u7684\u80fd\u529b\uff0c\u5e76\u5728 90% \u7684\u7b80\u5355\u548c\u4e2d\u7b49\u6c34\u5e73\u7684\u6218\u6597\u573a\u666f\u4e2d\u53d6\u5f97\u4e86\u6210\u529f\u3002<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.12889<\/span><\/section>\n
GitHub \u5730\u5740\uff1a<\/span><\/section>\n
https:\/\/varp-agent.github.io<\/span><\/section>\n

\"\"<\/p>\n

\u5c0f\u7ea2\u4e66\u63a8\u51fa StoryMaker\uff1a\u5b9e\u73b0\u201c\u6587\u751f\u56fe\u201d\u7684\u7279\u5f81\u6574\u4f53\u4e00\u81f4<\/span><\/section>\n
\u65e0\u9700\u989d\u5916\u5fae\u8c03\uff08Tuning-free\uff09\u7684\u4e2a\u6027\u5316\u56fe\u50cf\u751f\u6210\u65b9\u6cd5\u5728\u4fdd\u6301\u9762\u90e8\u4e00\u81f4\u6027\u65b9\u9762\u53d6\u5f97\u4e86\u5de8\u5927\u6210\u529f\u3002\u7136\u800c\uff0c\u5728\u6709\u591a\u4e2a\u89d2\u8272\u7684\u573a\u666f\u4e2d\uff0c\u7f3a\u4e4f\u6574\u4f53\u4e00\u81f4\u6027\u963b\u788d\u4e86\u8fd9\u4e9b\u65b9\u6cd5\u521b\u9020\u8fde\u8d2f\u53d9\u4e8b\u7684\u80fd\u529b\u3002<\/span><\/section>\n
\u5728\u8fd9\u9879\u5de5\u4f5c\u4e2d\uff0c\u5c0f\u7ea2\u4e66\u56e2\u961f\u63a8\u51fa\u4e86\u4e00\u79cd\u4e2a\u6027\u5316\u89e3\u51b3\u65b9\u6848\u2014\u2014StoryMaker\uff0c\u5b83\u4e0d\u4ec5\u80fd\u4fdd\u6301\u9762\u90e8\u7684\u4e00\u81f4\u6027\uff0c\u8fd8\u80fd\u4fdd\u6301\u670d\u88c5\u3001\u53d1\u578b\u548c\u8eab\u4f53\u7684\u4e00\u81f4\u6027\uff0c\u4ece\u800c\u901a\u8fc7\u4e00\u7cfb\u5217\u56fe\u50cf\u4fc3\u8fdb\u6545\u4e8b\u7684\u521b\u4f5c\u3002StoryMaker \u878d\u5408\u4e86\u57fa\u4e8e\u9762\u90e8\u8eab\u4efd\u7684\u6761\u4ef6\u548c\u88c1\u526a\u540e\u7684\u4eba\u7269\u56fe\u50cf\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u4ed6\u4eec\u4f7f\u7528\u4f4d\u7f6e\u611f\u77e5\u611f\u77e5\u5668\u91cd\u91c7\u6837\u5668\uff08<\/span>PPR<\/span>\uff09\u5c06\u9762\u90e8\u8eab\u4efd\u4fe1\u606f\u4e0e\u88c1\u526a\u540e\u7684\u4eba\u7269\u56fe\u50cf\u6574\u5408\u5728\u4e00\u8d77\uff0c\u4ece\u800c\u83b7\u5f97\u9c9c\u660e\u7684\u4eba\u7269\u7279\u5f81\u3002\u4e3a\u4e86\u9632\u6b62\u591a\u4e2a\u4eba\u7269\u548c\u80cc\u666f\u6df7\u6742\u5728\u4e00\u8d77\uff0c\u4ed6\u4eec\u4f7f\u7528\u5e26\u6709\u5206\u5272\u63a9\u7801\u7684 MSE \u635f\u5931\u5206\u522b\u9650\u5236\u4e0d\u540c\u4eba\u7269\u548c\u80cc\u666f\u7684\u4ea4\u53c9\u6ce8\u610f\u529b\u5f71\u54cd\u533a\u57df\u3002<\/span><\/span><\/section>\n
\u6b64\u5916\uff0c\u4ed6\u4eec\u4ee5\u59ff\u52bf\u4e3a\u6761\u4ef6\u8bad\u7ec3\u751f\u6210\u7f51\u7edc\uff0c\u4ece\u800c\u4fc3\u8fdb\u4e0e\u59ff\u52bf\u7684\u89e3\u8026\u3002\u4ed6\u4eec\u8fd8\u91c7\u7528\u4e86\u00a0<\/span>LoRA<\/span>\u00a0\u6765\u63d0\u9ad8\u4fdd\u771f\u5ea6\u548c\u8d28\u91cf\u3002<\/span><\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.12576<\/span><\/section>\n
GitHub \u5730\u5740\uff1a<\/span><\/section>\n
https:\/\/github.com\/RedAIGC\/StoryMaker<\/span><\/section>\n

\"\"<\/p>\n

\u5b57\u8282\u3001\u4e2d\u79d1\u9662\u56e2\u961f\u63a8\u51fa\u591a\u6a21\u6001\u6570\u5b66\u9884\u8bad\u7ec3\u6570\u636e\u96c6 InfiMM-WebMath-40B<\/span><\/section>\n
\u5728\u5927\u89c4\u6a21\u3001\u9ad8\u8d28\u91cf\u7684\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u9884\u8bad\u7ec3\u5bf9\u4e8e\u63d0\u9ad8\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u63a8\u7406\u80fd\u529b\u81f3\u5173\u91cd\u8981\uff0c\u5c24\u5176\u662f\u5728\u6570\u5b66\u7b49\u4e13\u4e1a\u9886\u57df\u3002\u5c3d\u7ba1\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff08MLLMs\uff09\u7684\u91cd\u8981\u6027\u5df2\u5f97\u5230\u516c\u8ba4\uff0c\u4f46\u8be5\u9886\u57df\u76ee\u524d\u4ecd\u7f3a\u4e4f\u4e13\u95e8\u9488\u5bf9\u6570\u5b66\u63a8\u7406\u7684\u5168\u9762\u5f00\u6e90\u9884\u8bad\u7ec3\u6570\u636e\u96c6\u3002<\/span><\/section>\n
\u4e3a\u4e86\u586b\u8865\u8fd9\u4e00\u7a7a\u767d\uff0c\u6765\u81ea\u5b57\u8282\u8df3\u52a8\u548c\u4e2d\u79d1\u9662\u7684\u7814\u7a76\u56e2\u961f\u63a8\u51fa\u4e86\u4e00\u4e2a\u9ad8\u8d28\u91cf\u7684\u4ea4\u9519\u56fe\u50cf-\u6587\u672c\u6587\u6863\u6570\u636e\u96c6\u2014\u2014InfiMM-WebMath-40B\u3002\u8be5\u6570\u636e\u96c6\u7531 2400 \u4e07\u4e2a\u7f51\u9875\u30018500 \u4e07\u4e2a\u76f8\u5173\u56fe\u7247 URL \u548c 400 \u4ebf\u4e2a\u6587\u672c token \u7ec4\u6210\u3002\u4e3a\u4e86\u8bc1\u660e InfiMM-WebMath-40B \u7684\u9c81\u68d2\u6027\uff0c\u4ed6\u4eec\u5728\u7eaf\u6587\u672c\u548c\u591a\u6a21\u6001\u73af\u5883\u4e2d\u8fdb\u884c\u4e86\u8bc4\u4f30\u3002\u5728\u7eaf\u6587\u672c\u57fa\u51c6\u4e0a\u7684\u8bc4\u4f30\u7ed3\u679c\u8868\u660e\uff0c\u5c3d\u7ba1\u53ea\u4f7f\u7528\u4e86 400 \u4ebf\u4e2a token\uff0c\u4f46\u6570\u636e\u96c6\u663e\u8457\u63d0\u9ad8\u4e86 1.3B \u53c2\u6570\u6a21\u578b\u7684\u6027\u80fd\uff0c\u5176\u7ed3\u679c\u53ef\u4e0e DeepSeekMath-1.3B \u76f8\u5ab2\u7f8e\uff0c\u540e\u8005\u5728\u76f8\u540c\u7684\u6a21\u578b\u89c4\u6a21\u4e0b\u4f7f\u7528\u4e86 1200 \u4ebf\u4e2a token\u3002<\/span><\/section>\n
\u5c3d\u7ba1\u5982\u6b64\uff0c\u968f\u7740\u591a\u6a21\u6001\u6570\u5b66\u9884\u8bad\u7ec3\u6570\u636e\u96c6\u7684\u5f15\u5165\uff0c\u4ed6\u4eec\u7684\u6a21\u578b\u5728\u591a\u6a21\u6001\u6570\u5b66\u57fa\u51c6\uff08\u5982 MathVerse \u548c We-Math\uff09\u4e0a\u521b\u9020\u4e86\u8fbe\u5230\u4e86\u5f00\u6e90\u6a21\u578b SOTA\u3002<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.12568<\/span><\/section>\n

\"\"<\/p>\n

\u7efc\u8ff0\uff1a\u8bed\u8a00\u3001\u8bed\u97f3\u548c\u89c6\u89c9\u4efb\u52a1\u4e2d\u7684\u504f\u597d\u5fae\u8c03<\/span><\/section>\n
\u504f\u597d\u5fae\u8c03\u662f\u4f7f\u6df1\u5ea6\u751f\u6210\u6a21\u578b\u4e0e\u4eba\u7c7b\u504f\u597d\u76f8\u4e00\u81f4\u7684\u5173\u952e\u8fc7\u7a0b\u3002\u5728\u8fd9\u9879\u5de5\u4f5c\u4e2d\uff0c\u6765\u81ea\u00a0<\/span>Capital One<\/span>\u00a0\u548c\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66\u7684\u7814\u7a76\u56e2\u961f\u5168\u9762\u6982\u8ff0\u4e86\u504f\u597d\u5fae\u8c03\u548c\u4eba\u7c7b\u53cd\u9988\u6574\u5408\u65b9\u9762\u7684\u6700\u65b0\u8fdb\u5c55\u3002\u5206\u4e3a\u4e09\u4e2a\u4e3b\u8981\u90e8\u5206\uff1a1\uff09\u5f15\u8a00\u548c\u524d\u8a00\uff1a\u4ecb\u7ecd\u5f3a\u5316\u5b66\u4e60\u6846\u67b6\u3001\u504f\u597d\u5fae\u8c03\u4efb\u52a1\u3001\u6a21\u578b\u548c\u5404\u79cd\u6a21\u5f0f\u7684\u6570\u636e\u96c6\uff1a\u8bed\u8a00\u3001\u8bed\u97f3\u548c\u89c6\u89c9\uff0c\u4ee5\u53ca\u4e0d\u540c\u7684\u7b56\u7565\u65b9\u6cd5\uff1b2\uff09\u6df1\u5165\u7814\u7a76\u6bcf\u79cd\u504f\u597d\u5fae\u8c03\u65b9\u6cd5\uff1a\u8be6\u7ec6\u5206\u6790\u504f\u597d\u5fae\u8c03\u4e2d\u4f7f\u7528\u7684\u65b9\u6cd5\uff1b3\uff09\u5e94\u7528\u3001\u8ba8\u8bba\u548c\u672a\u6765\u65b9\u5411\uff1a\u63a2\u8ba8\u504f\u597d\u5fae\u8c03\u5728\u4e0b\u6e38\u4efb\u52a1\u4e2d\u7684\u5e94\u7528\uff0c\u5305\u62ec\u4e0d\u540c\u6a21\u6001\u7684\u8bc4\u4f30\u65b9\u6cd5\uff0c\u4ee5\u53ca\u5bf9\u672a\u6765\u7814\u7a76\u65b9\u5411\u7684\u5c55\u671b\u3002<\/span><\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.11564<\/span><\/section>\n

\"\"<\/p>\n

\u5fae\u8f6f\u63d0\u51fa MoE \u8bad\u7ec3\u65b0\u65b9\u6cd5 GRIN<\/span><\/section>\n
\u6df7\u5408\u4e13\u5bb6\u6a21\u578b\uff08MoE\uff09\u901a\u8fc7\u4e13\u5bb6\u8def\u7531\u8fdb\u884c\u7a00\u758f\u8ba1\u7b97\uff0c\u53ea\u9009\u62e9\u6027\u5730\u6fc0\u6d3b\u4e00\u5c0f\u90e8\u5206\u4e13\u5bb6\u6a21\u5757\uff0c\u56e0\u6b64\u6bd4\u7a20\u5bc6\u6a21\u578b\u66f4\u80fd\u6709\u6548\u6269\u5c55\u3002\u7136\u800c\uff0c\u7a00\u758f\u8ba1\u7b97\u5bf9\u4f20\u7edf\u7684\u8bad\u7ec3\u65b9\u6cd5\u63d0\u51fa\u4e86\u6311\u6218\uff0c\u56e0\u4e3a\u79bb\u6563\u7684\u4e13\u5bb6\u8def\u7531\u4f1a\u963b\u788d\u6807\u51c6\u53cd\u5411\u4f20\u64ad\uff0c\u4ece\u800c\u963b\u788d\u57fa\u4e8e\u68af\u5ea6\u7684\u4f18\u5316\uff0c\u800c\u68af\u5ea6\u4f18\u5316\u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u91cd\u8981\u4e00\u73af\u3002<\/span><\/section>\n
\u4e3a\u4e86\u66f4\u597d\u5730\u53d1\u6325 MoE \u7684\u6269\u5c55\u80fd\u529b\uff0c\u5fae\u8f6f\u56e2\u961f\u63d0\u51fa\u4e86 GRIN\uff08GRadient-INformed MoE training\uff09\uff0c\u5b83\u5c06\u7a00\u758f\u68af\u5ea6\u4f30\u8ba1\u7528\u4e8e\u4e13\u5bb6\u8def\u7531\uff0c\u5e76\u914d\u7f6e\u6a21\u578b\u5e76\u884c\u6027\u4ee5\u907f\u514d token \u4e22\u5931\u3002\u5c06 GRIN \u5e94\u7528\u4e8e\u81ea\u56de\u5f52\u8bed\u8a00\u5efa\u6a21\uff0c\u4ed6\u4eec\u5f00\u53d1\u51fa\u4e86\u4e00\u4e2a top-2 16\u00d73.8B MoE \u6a21\u578b\u3002\u8fd9\u4e00\u6a21\u578b\u4ec5\u6709 6.6B \u6fc0\u6d3b\u53c2\u6570\uff0c\u5176\u6027\u80fd\u8d85\u8fc7\u4e86 7B \u7a20\u5bc6\u6a21\u578b\uff0c\u5e76\u4e0e\u5728\u76f8\u540c\u6570\u636e\u4e0a\u8bad\u7ec3\u7684 14B \u5bc6\u96c6\u6a21\u578b\u4e0d\u76f8\u4e0a\u4e0b\u3002<\/span><\/section>\n
\u5bf9\u4e0d\u540c\u4efb\u52a1\u7684\u5e7f\u6cdb\u8bc4\u4f30\u8868\u660e\uff0cGRIN \u6709\u6f5c\u529b\u63d0\u9ad8 MoE \u7684\u6548\u7387\uff0c\u5728 MMLU\u3001HellaSwag\u3001HumanEval \u548c MATH \u4e0a\u5206\u522b\u53d6\u5f97\u4e86 79.4\u300183.7\u300174.4 \u548c 58.9 \u7684\u5206\u6570\u3002<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.12136<\/span><\/section>\n

\"\"<\/p>\n

JourneyBench\uff1a\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u7684\u89c6\u89c9\u7406\u89e3\u8bc4\u4f30\u57fa\u51c6<\/span><\/section>\n
\u6700\u8fd1\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u53ea\u80fd\u4f9d\u9760\u80cc\u666f\u8bed\u8a00\u504f\u5dee\uff0c\u5728\u6d45\u5c42\u89c6\u89c9\u7406\u89e3\u7684\u57fa\u7840\u4e0a\u5b9e\u73b0\u826f\u597d\u7684\u6027\u80fd\u3002\u56e0\u6b64\uff0c\u5728\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4f18\u5f02\u5e76\u4e0d\u4e00\u5b9a\u4e0e\u89c6\u89c9\u7406\u89e3\u80fd\u529b\u5f3a\u6709\u5173\u3002<\/span><\/section>\n
\u6765\u81ea\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66\u3001\u5f17\u5409\u5c3c\u4e9a\u7406\u5de5\u5927\u5b66\u548c\u52a0\u5dde\u5927\u5b66\u6d1b\u6749\u77f6\u5206\u6821\u7684\u7814\u7a76\u56e2\u961f\uff0c\u63a8\u51fa\u4e86\u4e00\u4e2a\u7531\u4eba\u7c7b\u6807\u6ce8\u7684\u751f\u6210\u56fe\u50cf\u7684\u7efc\u5408\u57fa\u51c6\u2014\u2014JourneyBench\uff0c\u65e8\u5728\u8bc4\u4f30\u6a21\u578b\u5728\u4ee5\u4e0b\u4e94\u9879\u4efb\u52a1\u4e2d\u7684\u7ec6\u7c92\u5ea6\u591a\u6a21\u6001\u63a8\u7406\u80fd\u529b\uff1a\u4e92\u8865\u591a\u6a21\u6001\u601d\u7ef4\u94fe\u3001\u591a\u56fe\u50cf VQA\u3001\u865a\u6784\u56fe\u50cf\u63cf\u8ff0\u3001\u5e26\u6709\u5e7b\u89c9\u89e6\u53d1\u5668\u7684 VQA\uff0c\u4ee5\u53ca\u5e26\u6709\u7279\u5b9a\u6837\u672c\u5e72\u6270\u9879\u7684\u7ec6\u7c92\u5ea6\u68c0\u7d22\u3002\u4e0e\u73b0\u6709\u57fa\u51c6\u4e0d\u540c\u7684\u662f\uff0cJourneyBench \u660e\u786e\u8981\u6c42\u5728\u4e0d\u5bfb\u5e38\u7684\u60f3\u8c61\u573a\u666f\u4e2d\u8fdb\u884c\u7ec6\u7c92\u5ea6\u591a\u6a21\u6001\u63a8\u7406\uff0c\u800c\u5728\u8fd9\u4e9b\u573a\u666f\u4e2d\uff0c\u8bed\u8a00\u504f\u5dee\u548c\u6574\u4f53\u56fe\u50cf\u8981\u70b9\u662f\u4e0d\u591f\u7684\u3002<\/span><\/section>\n
\u4ed6\u4eec\u5728 JourneyBench \u4e0a\u5bf9 SOTA \u6a21\u578b\u8fdb\u884c\u4e86\u57fa\u51c6\u6d4b\u8bd5\uff0c\u5e76\u4ece\u591a\u4e2a\u7ec6\u7c92\u5ea6\u7ef4\u5ea6\u5bf9\u6027\u80fd\u8fdb\u884c\u4e86\u5206\u6790\u3002\u6240\u6709\u4e94\u9879\u4efb\u52a1\u7684\u7ed3\u679c\u8868\u660e\uff0c\u5373\u4f7f\u5bf9 SOTA \u6765\u8bf4\uff0cJourneyBench \u4e5f\u6781\u5177\u6311\u6218\u6027\uff0c\u8fd9\u8868\u660e\u6a21\u578b\u7684\u89c6\u89c9\u63a8\u7406\u80fd\u529b\u5e76\u4e0d\u50cf\u6700\u521d\u770b\u8d77\u6765\u90a3\u4e48\u5f3a\u3002<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.12953<\/span><\/section>\n

\"\"<\/p>\n

Promptriever\uff1a\u9996\u4e2a\u80fd\u591f\u50cf LM \u4e00\u6837\u8fdb\u884c\u63d0\u793a\u7684\u68c0\u7d22\u6a21\u578b<\/span><\/section>\n
\u7ecf\u8fc7\u6307\u4ee4\u5fae\u8c03\u7684\u8bed\u8a00\u6a21\u578b\uff08LM\uff09\u80fd\u591f\u54cd\u5e94\u6307\u4ee4\u6027\u547d\u4ee4\uff0c\u8fdb\u800c\u63d0\u4f9b\u6bd4\u57fa\u7840\u6a21\u578b\u66f4\u81ea\u7136\u7684\u7528\u6237\u754c\u9762\u3002\u5728\u8fd9\u9879\u5de5\u4f5c\u4e2d\uff0c\u6765\u81ea\u7ea6\u7ff0\u970d\u666e\u91d1\u65af\u5927\u5b66\u548c Samaya AI \u7684\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86\u9996\u4e2a\u80fd\u591f\u50cf LM \u4e00\u6837\u8fdb\u884c\u63d0\u793a\u7684\u68c0\u7d22\u6a21\u578b\u2014\u2014Promptriever\u3002\u4e3a\u4e86\u8bad\u7ec3 Promptriever\uff0c\u4ed6\u4eec\u4ece MS MARCO \u6536\u96c6\u5e76\u53d1\u5e03\u4e86\u4e00\u4e2a\u65b0\u7684\u5b9e\u4f8b\u7ea7\u6307\u4ee4\u8bad\u7ec3\u96c6\uff0c\u6db5\u76d6\u8fd1 50 \u4e07\u4e2a\u5b9e\u4f8b\u3002Promptriever \u4e0d\u4ec5\u5728\u6807\u51c6\u68c0\u7d22\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\uff0c\u800c\u4e14\u8fd8\u80fd\u8ddf\u968f\u6307\u4ee4\u3002<\/span><\/section>\n
\u4ed6\u4eec\u89c2\u5bdf\u5230\uff1a\uff081\uff09\u5728\u8ddf\u968f\u8be6\u7ec6\u7684\u76f8\u5173\u6027\u6307\u4ee4\u65b9\u9762\u53d6\u5f97\u4e86\u5de8\u5927\u8fdb\u6b65\uff08\u8fbe\u5230\u4e86 SoTA\uff09\uff08FollowIR \u4e0a +14.3 p-MRR \/ +3.1 nDCG\uff09\uff0c\uff082\uff09\u5bf9\u67e5\u8be2\u3001\u6307\u4ee4\u4e2d\u8bcd\u6c47\u9009\u62e9\/\u63aa\u8f9e\u7684\u9c81\u68d2\u6027\u663e\u8457\u63d0\u9ad8\uff08InstructIR \u4e0a +12.9 Robustness@10\uff09\uff0c\uff083\uff09\u80fd\u591f\u901a\u8fc7\u63d0\u793a\u6267\u884c\u8d85\u53c2\u6570\u641c\u7d22\uff0c\u4ece\u800c\u53ef\u9760\u5730\u63d0\u9ad8\u68c0\u7d22\u6027\u80fd\uff08BEIR \u4e0a\u5e73\u5747\u63d0\u9ad8 +1.4\uff09\u3002Promptriever \u8bc1\u660e\u4e86\u68c0\u7d22\u6a21\u578b\u53ef\u4ee5\u5728\u6bcf\u6b21\u67e5\u8be2\u7684\u57fa\u7840\u4e0a\u901a\u8fc7\u63d0\u793a\u8fdb\u884c\u63a7\u5236\uff0c\u4e3a\u4eca\u540e\u5c06 LM \u63d0\u793a\u6280\u672f\u4e0e\u4fe1\u606f\u68c0\u7d22\u76f8\u7ed3\u5408\u7684\u5de5\u4f5c\u5960\u5b9a\u4e86\u57fa\u7840\u3002<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/arxiv.org\/abs\/2409.11136<\/span><\/section>\n

\"\"<\/p>\n

\n
\n
\u7279\u522b\u8bf4\u660e\uff1a\u672c\u6587\u4ec5\u7528\u4e8e\u5b66\u672f\u4ea4\u6d41\uff0c\u5982\u6709\u4fb5\u6743\u8bf7\u540e\u53f0\u8054\u7cfb\u5c0f\u7f16\u5220\u9664\u3002<\/span><\/span><\/section>\n<\/section>\n<\/blockquote>\n
<\/section>\n<\/section>\n

 <\/p>\n<\/section>\n","protected":false},"excerpt":{"rendered":"

\u5927\u6a21\u578b\u73a9\u300a\u9ed1\u795e\u8bdd\uff1a\u609f\u7a7a\u300b\uff0c\u5b8c\u6210 90% \u7b80\u5355\u3001\u4e2d\u7b49\u6c34\u5e73\u6218\u6597 \u7ffb\u8bd1\u6280\u672f\u6559\u80b2\u4e0e\u7814\u7a76   2024\u5e7409\u6708 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[391],"tags":[],"class_list":["post-34144","post","type-post","status-publish","format-standard","hentry","category-391"],"_links":{"self":[{"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/posts\/34144","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/linguaresources.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=34144"}],"version-history":[{"count":1,"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/posts\/34144\/revisions"}],"predecessor-version":[{"id":34153,"href":"https:\/\/linguaresources.com\/index.php?rest_route=\/wp\/v2\/posts\/34144\/revisions\/34153"}],"wp:attachment":[{"href":"https:\/\/linguaresources.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=34144"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/linguaresources.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=34144"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/linguaresources.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=34144"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}