{"id":38793,"date":"2025-02-23T07:01:06","date_gmt":"2025-02-22T23:01:06","guid":{"rendered":"https:\/\/17aitech.com\/?p=38793"},"modified":"2025-02-23T07:01:06","modified_gmt":"2025-02-22T23:01:06","slug":"%e5%85%83%e8%b1%a1%e5%8f%91%e5%b8%83%e4%b8%ad%e5%9b%bd%e6%9c%80%e5%a4%a7moe%e5%bc%80%e6%ba%90%e5%a4%a7%e6%a8%a1%e5%9e%8b-%e8%90%bd%e5%9c%b0%e5%ba%94%e7%94%a8%e7%99%bb%e9%a1%b6%e6%b8%af%e5%8f%b0","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38793","title":{"rendered":"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c"},"content":{"rendered":"

\u5143\u8c61XVERSE\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u6a21\u578b\uff1aXVERSE-MoE-A36B\uff0c\u52a0\u901fAI\u5e94\u7528\u4f4e\u6210\u672c\u90e8\u7f72\uff0c\u5c06\u56fd\u4ea7\u5f00\u6e90\u63d0\u5347\u81f3\u56fd\u9645\u9886\u5148\u6c34\u5e73\u3002\u8be5\u6a21\u578b\u603b\u53c2\u6570255B\uff0c\u6fc0\u6d3b\u53c2\u657036B\uff0c\u8fbe\u5230100B\u6a21\u578b\u6027\u80fd\u7684\u300c\u8de8\u7ea7\u300d\u8dc3\u5347\uff0c\u540c\u65f6\u8bad\u7ec3\u65f6\u95f4\u51cf\u5c1130%\uff0c\u63a8\u7406\u6027\u80fd\u63d0\u5347100%\uff0c\u4f7f\u6bcftoken\u6210\u672c\u5927\u5e45\u4e0b\u964d\u3002<\/span><\/p>\n

\u5143\u8c61\u300c\u9ad8\u6027\u80fd\u5168\u5bb6\u6876\u300d\u7cfb\u5217\u5168\u90e8\u5f00\u6e90\uff0c\u65e0\u6761\u4ef6\u514d\u8d39\u5546\u7528\uff0c\u8ba9\u6d77\u91cf\u4e2d\u5c0f\u4f01\u4e1a\u3001\u7814\u7a76\u8005\u548c\u5f00\u53d1\u8005\u80fd\u6309\u9700\u9009\u62e9\u3002<\/p>\n

<\/a>\u00a0MoE\uff08Mixture of Experts\uff09\u662f\u4e1a\u754c\u6700\u524d\u6cbf\u7684\u6df7\u5408\u4e13\u5bb6\u6a21\u578b\u67b6\u6784 \uff0c\u5c06\u591a\u4e2a\u7ec6\u5206\u9886\u57df\u7684\u4e13\u5bb6\u6a21\u578b\u7ec4\u5408\u6210\u4e00\u4e2a\u8d85\u7ea7\u6a21\u578b\uff0c\u6253\u7834\u4e86\u4f20\u7edf\u6269\u5c55\u5b9a\u5f8b\uff08Scaling Law\uff09\u7684\u5c40\u9650\uff0c\u53ef\u5728\u6269\u5927\u6a21\u578b\u89c4\u6a21\u65f6\uff0c\u4e0d\u663e\u8457\u589e\u52a0\u8bad\u7ec3\u548c\u63a8\u7406\u7684\u8ba1\u7b97\u6210\u672c\uff0c\u4fdd\u6301\u6a21\u578b\u6027\u80fd\u6700\u5927\u5316\u3002\u51fa\u4e8e\u8fd9\u4e2a\u539f\u56e0\uff0c\u884c\u4e1a\u524d\u6cbf\u6a21\u578b\u5305\u62ec\u8c37\u6b4cGemini-1.5\u3001OpenAI\u7684GPT-4\u00a0\u3001\u9a6c\u65af\u514b\u65d7\u4e0bxAI\u516c\u53f8\u7684Grok\u7b49\u5927\u6a21\u578b\u90fd\u4f7f\u7528\u4e86\u00a0MoE\u3002<\/span><\/p>\n

\u5728\u591a\u4e2a\u6743\u5a01\u8bc4\u6d4b\u4e2d\uff0c\u5143\u8c61MoE\u6548\u679c\u5927\u5e45\u8d85\u8d8a\u591a\u4e2a\u540c\u7c7b\u6a21\u578b\uff0c\u5305\u62ec\u56fd\u5185\u5343\u4ebfMoE\u6a21\u578b\u00a0Skywork-MoE\u3001\u4f20\u7edfMoE\u9738\u4e3bMixtral-8x22B\u00a0\u4ee5\u53ca3140\u4ebf\u53c2\u6570\u7684MoE\u5f00\u6e90\u6a21\u578bGrok-1-A86B\u7b49\u3002\u00a0<\/p>\n

<\/a>\u00a0<\/p>\n

\u514d\u8d39\u4e0b\u8f7d\u5927\u6a21\u578b<\/p>\n

Hugging Face\uff1ahttps:\/\/huggingface.co\/xverse\/XVERSE-MoE-A36B<\/p>\n

\u9b54\u642d\uff1ahttps:\/\/modelscope.cn\/models\/xverse\/XVERSE-MoE-A36B<\/p>\n

Github\uff1ahttps:\/\/github.com\/xverse-ai\/XVERSE-MoE-A36B<\/p>\n

\u5b98\u7f51\uff1achat.xverse.cn<\/p>\n

\u00a0<\/p>\n

\u843d\u5730\u5e94\u7528\u597d\u4e14\u7701\u00a0\u767b\u9876\u6e2f\u53f0\u5a31\u4e50\u5e94\u7528\u699c<\/p>\n

<\/p>\n

\u5143\u8c61\u6b64\u6b21\u5f00\u6e90\uff0c\u4e0d\u4ec5\u586b\u8865\u56fd\u5185\u7a7a\u767d\uff0c\u4e5f\u5728\u5546\u4e1a\u5e94\u7528\u4e0a\u66f4\u8fdb\u4e00\u6b65\u3002<\/p>\n

\u5143\u8c61\u57fa\u4e8eMoE\u6a21\u578b\u81ea\u4e3b\u7814\u53d1\u7684AI\u89d2\u8272\u626e\u6f14\u4e0e\u4e92\u52a8\u7f51\u6587APP Saylo\uff0c\u901a\u8fc7\u903c\u771f\u7684AI\u89d2\u8272\u626e\u6f14\u548c\u6709\u8da3\u7684\u5f00\u653e\u5267\u60c5\uff0c\u706b\u904d\u6e2f\u53f0\uff0c\u4e0b\u8f7d\u91cf\u5728\u4e2d\u56fd\u53f0\u6e7e\u548c\u9999\u6e2f\u5a31\u4e50\u699c\u5206\u522b\u4f4d\u5217\u7b2c\u4e00\u548c\u7b2c\u4e09\u3002<\/p>\n

MoE\u8bad\u7ec3\u8303\u5f0f\u5177\u6709\u300c\u66f4\u9ad8\u6027\u80fd\u3001\u66f4\u4f4e\u6210\u672c\u300d\u4f18\u52bf\uff0c\u5143\u8c61\u5728\u901a\u7528\u9884\u8bad\u7ec3\u57fa\u7840\u4e0a\uff0c\u4f7f\u7528\u6d77\u91cf\u5267\u672c\u6570\u636e\u300c\u7ee7\u7eed\u9884\u8bad\u7ec3\u300d\uff08Continue Pre-training\uff09\uff0c\u5e76\u4e0e\u4f20\u7edfSFT\uff08\u76d1\u7763\u5fae\u8c03\uff09\u6216RLHF\uff08\u57fa\u4e8e\u4eba\u7c7b\u53cd\u9988\u7684\u5f3a\u5316\u5b66\u4e60\uff09\u4e0d\u540c\uff0c\u91c7\u7528\u4e86\u5927\u89c4\u6a21\u8bed\u6599\u77e5\u8bc6\u6ce8\u5165\uff0c\u8ba9\u6a21\u578b\u65e2\u4fdd\u6301\u4e86\u5f3a\u5927\u7684\u901a\u7528\u8bed\u8a00\u7406\u89e3\u80fd\u529b\uff0c\u53c8\u5927\u5e45\u63d0\u5347\u300c\u5267\u672c\u300d\u8fd9\u4e00\u7279\u5b9a\u5e94\u7528\u9886\u57df\u7684\u8868\u73b0\u3002\u00a0\u00a0<\/span><\/a><\/p>\n

<\/p>\n

\u9ad8\u6027\u80fd\u300c\u5f00\u6e90\u6807\u6746\u300d<\/p>\n

\u00a0<\/p>\n

\u5143\u8c61\u662f\u56fd\u5185\u9886\u5148\u7684AI\u4e0e3D\u516c\u53f8\uff0c\u79c9\u6301\u300c\u901a\u7528\u4eba\u5de5\u667a\u80fd\u00a0AGI\u300d\u4fe1\u4ef0\uff0c\u6301\u7eed\u6253\u9020\u300c\u9ad8\u6027\u80fd\u5f00\u6e90\u5168\u5bb6\u6876\u300d\uff0c\u4e0d\u4ec5\u586b\u8865\u56fd\u4ea7\u5f00\u6e90\u7a7a\u767d\uff0c\u66f4\u5c06\u5176\u63a8\u5411\u4e86\u56fd\u9645\u9886\u5148\u6c34\u5e73\u3002<\/p>\n

<\/a><\/p>\n

2023\u5e7411\u6708\uff0c\u6b64\u524d\u56fd\u5185\u5927\u90e8\u5206\u5f00\u6e90\u53c2\u6570\u591a\u57287B\u523013B\uff0c\u800c\u884c\u4e1a\u5171\u8bc6\u662f\u6a21\u578b\u8fbe\u523050\u523060B\u53c2\u6570\u95e8\u69db\uff0c\u5927\u6a21\u578b\u624d\u80fd\u201c\u667a\u80fd\u6d8c\u73b0\u201d\uff0c\u751f\u6001\u4e9f\u9700\u201c\u5927\u201d\u6a21\u578b\u65f6\uff0c\u5143\u8c61\u7387\u5148\u5f00\u6e90\u4e86XVERSE-65B\uff0c\u662f\u5f53\u65f6\u4e2d\u56fd\u6700\u5927\u53c2\u6570\u5f00\u6e90\u3002\u00a0<\/p>\n

<\/a><\/p>\n

2024\u5e741\u6708\uff0c\u5143\u8c61\u53c8\u5f00\u6e90\u5168\u7403\u6700\u957f\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u6a21\u578b\uff0c\u652f\u6301\u8f93\u516525\u4e07\u6c49\u5b57\uff0c\u8fd8\u9644\u624b\u628a\u624b\u8bad\u7ec3\u6559\u7a0b\uff0c\u8ba9\u5927\u6a21\u578b\u5e94\u7528\u4e00\u4e3e\u8fdb\u5165\u201c\u957f\u6587\u672c\u65f6\u4ee3\u201d\u3002\u00a0<\/p>\n

<\/a><\/p>\n

\u6b64\u6b21\u56fd\u5185\u6700\u5927\u53c2\u6570MoE\u5f00\u6e90\uff0c\u53c8\u662f\u7ed9\u751f\u6001\u8d21\u732e\u4e86\u4e00\u4e2a\u52a9\u63a8\u4f4e\u6210\u672cAI\u5e94\u7528\u5229\u5668\u3002<\/p>\n

\u00a0<\/p>\n

\u5f15\u9886\u6587\u5a31\u5e94\u7528<\/p>\n

\u00a0<\/p>\n

\u501f\u52a9\u5728AI\u548c3D\u9886\u57df\u7684\u5ba2\u6237\u79ef\u7d2f\uff0c\u5143\u8c61\u4e5f\u8fc5\u901f\u5c06\u5927\u6a21\u578b\u63a8\u5411\u5546\u7528\u3002<\/p>\n

2023\u5e7411\u6708\uff0c\u5143\u8c61\u6210\u4e3a\u5168\u56fd\u6700\u65e9\u4e00\u6279\u3001\u5e7f\u4e1c\u7701\u524d\u4e94\u83b7\u5f97\u300a\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u80fd\u670d\u52a1\u7ba1\u7406\u6682\u884c\u529e\u6cd5\u300b\u56fd\u5bb6\u5907\u6848\u7684\u5927\u6a21\u578b\uff0c\u5177\u5907\u5411\u5168\u793e\u4f1a\u5f00\u653e\u7684\u4ea7\u54c1\u80fd\u529b\u3002\u00a0<\/p>\n

\u800c\u5728\u66f4\u65e9\u768410\u6708\uff0c\u5143\u8c61\u4e0e\u817e\u8baf\u97f3\u4e50\u8054\u5408\u63a8\u51falyraXVERSE\u52a0\u901f\u5927\u6a21\u578b\uff0c\u5e76\u501f\u52a9\u8be5\u6280\u672f\u5168\u9762\u5347\u7ea7\u97f3\u4e50\u52a9\u624b\u201cAI\u5c0f\u7434\u201d\u7684\u95ee\u7b54\u3001\u804a\u5929\u4e0e\u521b\u4f5c\u80fd\u529b\uff0c\u8ba9\u5979\u60c5\u5546\u4e0e\u667a\u5546\u53cc\u9ad8\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u4e2a\u6027\u5316\u3001\u66f4\u6df1\u5165\u3001\u966a\u4f34\u611f\u5341\u8db3\u7684\u97f3\u4e50\u4e92\u52a8\u4f53\u9a8c\u3002\u00a0<\/p>\n

<\/a><\/p>\n

\u5143\u8c61\u5927\u6a21\u578b\u9646\u7eed\u4e0eQQ\u97f3\u4e50\u3001\u864e\u7259\u76f4\u64ad\u3001\u5168\u6c11K\u6b4c\u3001\u817e\u8baf\u4e91\u7b49\u6df1\u5ea6\u5408\u4f5c\u4e0e\u5e94\u7528\u63a2\u7d22\uff0c\u4e3a\u6587\u5316\u3001\u5a31\u4e50\u3001\u65c5\u6e38\u3001\u91d1\u878d\u9886\u57df\u6253\u9020\u521b\u65b0\u9886\u5148\u7684\u7528\u6237\u4f53\u9a8c\u3002\u00a0\u00a0<\/span><\/a><\/p>\n

<\/p>\n

MoE\u6280\u672f\u81ea\u7814\u4e0e\u521b\u65b0<\/p>\n

\u00a0<\/p>\n

MoE\u662f\u76ee\u524d\u4e1a\u754c\u6700\u524d\u6cbf\u7684\u6a21\u578b\u6846\u67b6\uff0c\u7531\u4e8e\u6280\u672f\u8f83\u65b0\uff0c\u56fd\u5185\u5f00\u6e90\u6a21\u578b\u6216\u5b66\u672f\u7814\u7a76\u5c1a\u672a\u666e\u53ca\u3002\u5143\u8c61\u81ea\u7814MoE\u7684\u9ad8\u6548\u8bad\u7ec3\u548c\u63a8\u7406\u6846\u67b6\uff0c\u5e76\u6301\u7eed\u63a8\u52a8\u6280\u672f\u521b\u65b0\u3002<\/p>\n

2024\u5e744\u6708\u63a8\u51fa\u7684XVERSE-MoE-A4.2B\u4e2d\uff0c\u5143\u8c61\u63a8\u52a8MoE\u4e13\u5bb6\u67b6\u6784\u9769\u65b0\u3002\u4e0e\u4f20\u7edfMoE\uff08\u5982Mixtral 8x7B\uff09\u5c06\u6bcf\u4e2a\u4e13\u5bb6\u5927\u5c0f\u7b49\u540c\u4e8e\u6807\u51c6FFN\u4e0d\u540c\uff0c\u5143\u8c61\u91c7\u7528\u66f4\u7ec6\u7c92\u5ea6\u7684\u4e13\u5bb6\u8bbe\u8ba1\uff0c\u6bcf\u4e2a\u4e13\u5bb6\u5927\u5c0f\u4ec5\u4e3a\u6807\u51c6FFN\u7684\u56db\u5206\u4e4b\u4e00\uff0c\u63d0\u9ad8\u4e86\u6a21\u578b\u7075\u6d3b\u6027\u4e0e\u6027\u80fd\uff1b\u8fd8\u5c06\u4e13\u5bb6\u5206\u4e3a\u5171\u4eab\u4e13\u5bb6\uff08Shared Expert\uff09\u548c\u975e\u5171\u4eab\u4e13\u5bb6\uff08Non-shared Expert\uff09\u4e24\u7c7b\u3002\u5171\u4eab\u4e13\u5bb6\u5728\u8ba1\u7b97\u8fc7\u7a0b\u4e2d\u59cb\u7ec8\u4fdd\u6301\u6fc0\u6d3b\u72b6\u6001\uff0c\u800c\u975e\u5171\u4eab\u4e13\u5bb6\u5219\u6839\u636e\u9700\u8981\u9009\u62e9\u6027\u6fc0\u6d3b\u3002\u8fd9\u79cd\u8bbe\u8ba1\u6709\u5229\u4e8e\u5c06\u901a\u7528\u77e5\u8bc6\u538b\u7f29\u81f3\u5171\u4eab\u4e13\u5bb6\u53c2\u6570\u4e2d\uff0c\u51cf\u5c11\u975e\u5171\u4eab\u4e13\u5bb6\u53c2\u6570\u95f4\u7684\u77e5\u8bc6\u5197\u4f59\u3002\u00a0<\/p>\n

<\/a><\/p>\n

\u6b64\u6b21\u63a8\u51faXVERSE-MoE-A36B\uff0c\u7ee7\u7eed\u5728MoE\u6548\u7387\u548c\u6548\u679c\u65b9\u9762\u8fdb\u884c\u6280\u672f\u521b\u65b0\u3002<\/p>\n

\uff081\uff09\u6548\u7387\u65b9\u9762<\/p>\n

MoE\u67b6\u6784\u4e0e4D\u62d3\u6251\u8bbe\u8ba1\uff1aMoE\u67b6\u6784\u7684\u5173\u952e\u7279\u6027\u662f\u7531\u591a\u4e2a\u4e13\u5bb6\u7ec4\u6210\u3002\u7531\u4e8e\u4e13\u5bb6\u4e4b\u95f4\u9700\u8981\u5927\u91cf\u7684\u4fe1\u606f\u4ea4\u6362\uff0c\u901a\u4fe1\u8d1f\u62c5\u6781\u91cd\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u6211\u4eec\u91c7\u7528\u4e864D\u62d3\u6251\u67b6\u6784\uff0c\u5e73\u8861\u4e86\u901a\u4fe1\u3001\u663e\u5b58\u548c\u8ba1\u7b97\u8d44\u6e90\u7684\u5206\u914d\u3002\u8fd9\u79cd\u8bbe\u8ba1\u4f18\u5316\u4e86\u8ba1\u7b97\u8282\u70b9\u4e4b\u95f4\u7684\u901a\u4fe1\u8def\u5f84\uff0c\u63d0\u9ad8\u4e86\u6574\u4f53\u8ba1\u7b97\u6548\u7387\u3002<\/p>\n

\u4e13\u5bb6\u8def\u7531\u4e0e\u9884\u4e22\u5f03\u7b56\u7565\uff1aMoE\u7684\u53e6\u4e00\u4e2a\u7279\u70b9\u662f\u201c\u4e13\u5bb6\u8def\u7531\u673a\u5236\u201d\uff0c\u5373\u9700\u8981\u5bf9\u4e0d\u540c\u7684\u8f93\u5165\u8fdb\u884c\u5206\u914d\uff0c\u5e76\u4e22\u5f03\u4e00\u4e9b\u8d85\u51fa\u4e13\u5bb6\u8ba1\u7b97\u5bb9\u91cf\u7684\u5197\u4f59\u6570\u636e\u3002\u4e3a\u6b64\u56e2\u961f\u8bbe\u8ba1\u4e00\u5957\u9884\u4e22\u5f03\u7b56\u7565\uff0c\u51cf\u5c11\u4e0d\u5fc5\u8981\u7684\u8ba1\u7b97\u548c\u4f20\u8f93\u3002\u540c\u65f6\u5728\u8ba1\u7b97\u6d41\u7a0b\u4e2d\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u7b97\u5b50\u878d\u5408\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u578b\u7684\u8bad\u7ec3\u6027\u80fd\u3002<\/p>\n

\u901a\u4fe1\u4e0e\u8ba1\u7b97\u91cd\u53e0\uff1a\u7531\u4e8eMoE\u67b6\u6784\u7684\u4e13\u5bb6\u4e4b\u95f4\u9700\u8981\u5927\u91cf\u901a\u4fe1\uff0c\u4f1a\u5f71\u54cd\u6574\u4f53\u8ba1\u7b97\u6548\u7387\u3002\u4e3a\u6b64\u56e2\u961f\u8bbe\u8ba1\u4e86\u201c\u591a\u7ef4\u5ea6\u7684\u901a\u4fe1\u4e0e\u8ba1\u7b97\u91cd\u53e0\u201d\u673a\u5236\uff0c\u5373\u5728\u8fdb\u884c\u53c2\u6570\u901a\u4fe1\u7684\u540c\u65f6\uff0c\u6700\u5927\u6bd4\u4f8b\u5e76\u884c\u5730\u6267\u884c\u8ba1\u7b97\u4efb\u52a1\uff0c\u4ece\u800c\u51cf\u5c11\u901a\u4fe1\u7b49\u5f85\u65f6\u95f4\u3002<\/p>\n

\uff082\uff09\u6548\u679c\u65b9\u9762<\/p>\n

\u4e13\u5bb6\u6743\u91cd\uff1aMoE\u00a0\u4e2d\u7684\u4e13\u5bb6\u603b\u6570\u4e3a\u00a0N\u00a0\uff0c\u6bcf\u4e2a\u00a0token\u00a0\u4f1a\u9009\u62e9\u00a0topK\u00a0\u4e2a\u4e13\u5bb6\u53c2\u4e0e\u540e\u7eed\u7684\u8ba1\u7b97\uff0c\u7531\u4e8e\u4e13\u5bb6\u5bb9\u91cf\u7684\u9650\u5236\uff0c\u6bcf\u4e2a\u00a0token\u00a0\u5b9e\u9645\u9009\u62e9\u5230\u7684\u4e13\u5bb6\u6570\u4e3a\u00a0M\uff0cM<=K<\/p>\n

\u5b9e\u9a8c1\uff1a\u6743\u91cd\u5728\u00a0topM\u00a0\u8303\u56f4\u5185\u5f52\u4e00\u5316<\/p>\n

\u5b9e\u9a8c2\uff1a\u6743\u91cd\u5728\u00a0topK\u00a0\u8303\u56f4\u5185\u5f52\u4e00\u5316<\/p>\n

\u5b9e\u9a8c3\uff1a\u6743\u91cd\u5728\u00a0topN\u00a0\u8303\u56f4\u5185\u5f52\u4e00\u5316<\/p>\n

\u5b9e\u9a8c4\uff1a\u6743\u91cd\u90fd\u4e3a\u00a01\u00a0<\/p>\n

<\/a><\/p>\n

\u5bf9\u6bd4\u5b9e\u9a8c\u7ed3\u679c<\/p>\n

\u4e3e\u4f8b\u8bf4\u660e\uff0c\u5047\u8bbeN=8\uff0cK=4\uff0cM=3\uff082\u53f7\u4e13\u5bb6\u4e0atoken\u88ab\u4e22\u5f03\uff09\uff0c\u4e0d\u540c\u4e13\u5bb6\u6743\u91cd\u7684\u8ba1\u7b97\u65b9\u5f0f\u6240\u5f97\u7684\u6743\u91cd\u5982\u4e0b\u56fe\uff1a\u00a0<\/p>\n

<\/a><\/p>\n

\u6570\u636e\u52a8\u6001\u5207\u6362\uff1a\u5143\u8c61\u4ee5\u5f80\u5f00\u6e90\u7684\u6a21\u578b\uff0c\u5f80\u5f80\u5728\u8bad\u7ec3\u524d\u5c31\u9501\u5b9a\u4e86\u8bad\u7ec3\u6570\u636e\u96c6\uff0c\u5e76\u5728\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4fdd\u6301\u4e0d\u53d8\u3002\u8fd9\u79cd\u505a\u6cd5\u867d\u7136\u7b80\u5355\uff0c\u4f46\u4f1a\u53d7\u5236\u4e8e\u521d\u59cb\u6570\u636e\u7684\u8d28\u91cf\u548c\u8986\u76d6\u9762\u3002\u6b64\u6b21MoE\u6a21\u578b\u7684\u8bad\u7ec3\u501f\u9274\u4e86”\u8bfe\u7a0b\u5b66\u4e60”\u7406\u5ff5\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5b9e\u73b0\u4e86\u52a8\u6001\u6570\u636e\u5207\u6362\uff0c\u5728\u4e0d\u540c\u9636\u6bb5\u591a\u6b21\u5f15\u5165\u65b0\u5904\u7406\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff0c\u5e76\u52a8\u6001\u8c03\u6574\u6570\u636e\u91c7\u6837\u6bd4\u4f8b\u3002<\/p>\n

\u8fd9\u8ba9\u6a21\u578b\u4e0d\u518d\u88ab\u521d\u59cb\u8bed\u6599\u96c6\u6240\u9650\u5236\uff0c\u800c\u662f\u80fd\u591f\u6301\u7eed\u5b66\u4e60\u65b0\u5f15\u5165\u7684\u9ad8\u8d28\u91cf\u6570\u636e\uff0c\u63d0\u5347\u4e86\u8bed\u6599\u8986\u76d6\u9762\u548c\u6cdb\u5316\u80fd\u529b\u3002\u540c\u65f6\u901a\u8fc7\u8c03\u6574\u91c7\u6837\u6bd4\u4f8b\uff0c\u4e5f\u6709\u52a9\u4e8e\u5e73\u8861\u4e0d\u540c\u6570\u636e\u6e90\u5bf9\u6a21\u578b\u6027\u80fd\u7684\u5f71\u54cd\u3002\u00a0<\/p>\n

<\/a><\/p>\n

\u4e0d\u540c\u6570\u636e\u7248\u672c\u7684\u6548\u679c\u66f2\u7ebf\u56fe<\/p>\n

\u5b66\u4e60\u7387\u8c03\u5ea6\u7b56\u7565\uff08LR Scheduler\uff09\uff1a\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u52a8\u6001\u5207\u6362\u6570\u636e\u96c6\uff0c\u867d\u6709\u52a9\u4e8e\u6301\u7eed\u5f15\u5165\u65b0\u77e5\u8bc6\uff0c\u4f46\u4e5f\u7ed9\u6a21\u578b\u5e26\u6765\u4e86\u65b0\u7684\u9002\u5e94\u6311\u6218\u3002\u4e3a\u4e86\u786e\u4fdd\u6a21\u578b\u80fd\u5feb\u901f\u4e14\u5145\u5206\u5730\u5b66\u4e60\u65b0\u8fdb\u6570\u636e\uff0c\u56e2\u961f\u5bf9\u5b66\u4e60\u7387\u8c03\u5ea6\u5668\u8fdb\u884c\u4e86\u4f18\u5316\u8c03\u6574\uff0c\u5728\u6bcf\u6b21\u6570\u636e\u5207\u6362\u65f6\u4f1a\u6839\u636e\u6a21\u578b\u6536\u655b\u72b6\u6001\uff0c\u76f8\u5e94\u8c03\u6574\u5b66\u4e60\u7387\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u8fd9\u4e00\u7b56\u7565\u6709\u6548\u63d0\u5347\u4e86\u6a21\u578b\u5728\u6570\u636e\u5207\u6362\u540e\u7684\u5b66\u4e60\u901f\u5ea6\u548c\u6574\u4f53\u8bad\u7ec3\u6548\u679c\u3002<\/p>\n

\u4e0b\u56fe\u662f\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u00a0MMLU\u3001HumanEval\u00a0\u4e24\u4e2a\u8bc4\u6d4b\u6570\u636e\u96c6\u7684\u6548\u679c\u66f2\u7ebf\u56fe\u3002<\/p>\n

<\/a><\/p>\n

\u8bad\u7ec3\u8fc7\u7a0b\u4e2dMMLU\u3001HumanEval\u7684\u6027\u80fd\u66f2\u7ebf\u6301\u7eed\u62d4\u9ad8<\/p>\n

\u901a\u8fc7\u8bbe\u8ba1\u4e0e\u4f18\u5316\uff0c\u5143\u8c61MoE\u6a21\u578b\u4e0e\u5176Dense\u6a21\u578bXVERSE-65B-2\u76f8\u6bd4\uff0c\u8bad\u7ec3\u65f6\u95f4\u51cf\u5c1130%\u3001\u63a8\u7406\u6027\u80fd\u63d0\u5347100%\uff0c\u6a21\u578b\u6548\u679c\u66f4\u4f73\u3002\u96f7\u5cf0\u7f51\u96f7\u5cf0\u7f51(\u516c\u4f17\u53f7\uff1a\u96f7\u5cf0\u7f51)<\/span><\/span><\/p>\n

<\/p>\n

<\/p>\n

<\/p>\n

\u96f7\u5cf0\u7f51\u539f\u521b\u6587\u7ae0\uff0c\u672a\u7ecf\u6388\u6743\u7981\u6b62\u8f6c\u8f7d\u3002\u8be6\u60c5\u89c1\u8f6c\u8f7d\u987b\u77e5<\/a>\u3002<\/p>\n

\n
\n <\/a>
\n <\/a>\n <\/p>\n<\/p>\n

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u5143\u8c61XVERSE\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u6a21 […]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","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":"","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":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","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":[27],"tags":[68],"class_list":["post-38793","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"\n\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c - \u4e00\u8d77AI\u6280\u672f<\/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:\/\/17aitech.com\/?p=38793\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c\",\"datePublished\":\"2025-02-22T23:01:06+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793\"},\"wordCount\":137,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38793#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=38793\",\"name\":\"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png\",\"datePublished\":\"2025-02-22T23:01:06+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=38793\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=38793#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\",\"url\":\"https:\\\/\\\/17aitech.com\\\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/17aitech.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"url\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"contentUrl\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c - \u4e00\u8d77AI\u6280\u672f","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:\/\/17aitech.com\/?p=38793","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=38793#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=38793"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c","datePublished":"2025-02-22T23:01:06+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=38793"},"wordCount":137,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=38793#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=38793#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38793","url":"https:\/\/17aitech.com\/?p=38793","name":"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38793#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38793#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png","datePublished":"2025-02-22T23:01:06+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38793#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38793"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38793#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-2ebd24b96c9d560538c6bd1e27e93aaf.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38793#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u5143\u8c61\u53d1\u5e03\u4e2d\u56fd\u6700\u5927MoE\u5f00\u6e90\u5927\u6a21\u578b \u843d\u5730\u5e94\u7528\u767b\u9876\u6e2f\u53f0\u699c"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38793","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=38793"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38793\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38793"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38793"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38793"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}