{"id":483,"date":"2024-11-20T16:58:17","date_gmt":"2024-11-20T08:58:17","guid":{"rendered":"https:\/\/www.bosombot.com\/?p=483"},"modified":"2024-11-20T17:07:04","modified_gmt":"2024-11-20T09:07:04","slug":"%e9%9c%87%e6%92%bc%e5%8f%91%e7%8e%b0%ef%bc%9aairllm-%e8%b5%8b%e8%83%bd-70b-%e5%a4%a7%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/www.bosombot.com\/?p=483","title":{"rendered":"\u9707\u64bc\u53d1\u73b0\uff1aAirLLM \u8d4b\u80fd 70B \u5927\u6a21\u578b"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_7\"><\/a>\u524d\u8a00<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728<a href=\"https:\/\/so.csdn.net\/so\/search?q=%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD&amp;spm=1001.2101.3001.7020\" target=\"_blank\" rel=\"noreferrer noopener\">\u4eba\u5de5\u667a\u80fd<\/a>\u84ec\u52c3\u53d1\u5c55\u7684\u6d6a\u6f6e\u4e2d\uff0c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u65e0\u7591\u662f\u6700\u8000\u773c\u7684\u660e\u73e0\u3002\u5b83\u4eec\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u5185\u5bb9\u751f\u6210\u3001\u667a\u80fd\u95ee\u7b54\u7b49\u4f17\u591a\u9886\u57df\u5c55\u73b0\u51fa\u4e86\u60ca\u4eba\u7684\u80fd\u529b\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u5f3a\u5927\u7684\u6a21\u578b\u5728\u63a8\u7406\u9636\u6bb5\u5f80\u5f80\u5bf9\u8ba1\u7b97\u8d44\u6e90\u548c\u5185\u5b58\u6709\u7740\u6781\u9ad8\u7684\u8981\u6c42\uff0c\u8fd9\u6210\u4e3a\u4e86\u5176\u5e7f\u6cdb\u5e94\u7528\u7684\u4e00\u5927\u74f6\u9888\u3002\u8bb8\u591a\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u53d7\u9650\u4e8e\u786c\u4ef6\u6761\u4ef6\uff0c\u65e0\u6cd5\u5145\u5206\u6316\u6398\u5927\u6a21\u578b\u7684\u6f5c\u529b\u3002\u5728\u6b64\u80cc\u666f\u4e0b\uff0c<strong>AirLLM\u5e94\u8fd0\u800c\u751f\uff0c\u5b83\u662f\u4e00\u6b3e\u4e13\u95e8\u9488\u5bf9\u5927\u578b\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u9636\u6bb5\u5185\u5b58\u4f7f\u7528\u4f18\u5316\u7684\u5f00\u6e90\u5de5\u5177\u5305\uff0c\u4e3a\u5728\u8d44\u6e90\u53d7\u9650\u73af\u5883\u4e0b\u5b9e\u73b0\u5927\u6a21\u578b\u7684\u9ad8\u6548\u63a8\u7406\u5f00\u8f9f\u4e86\u65b0\u7684\u9014\u5f84\u3002<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" src=\"https:\/\/greatbay.org.cn\/wp-content\/uploads\/2024\/11\/image-2.png\"  class=\"wp-image-348\" title=\"\u9707\u64bc\u53d1\u73b0\uff1aAirLLM \u8d4b\u80fd 70B \u5927\u6a21\u578b\u63d2\u56fe\" alt=\"\u9707\u64bc\u53d1\u73b0\uff1aAirLLM \u8d4b\u80fd 70B \u5927\u6a21\u578b\u63d2\u56fe\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_12\"><\/a>\u4e00\u3001\u6280\u672f\u4eae\u70b9<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">1\u3001<strong>\u5206\u5c42\u63a8\u7406<\/strong><br>AirLLM\u7684\u5206\u5c42\u63a8\u7406\u57fa\u4e8e\u5206\u800c\u6cbb\u4e4b\u7684\u7406\u5ff5\u3002<code>\u5927\u578b\u8bed\u8a00\u6a21\u578b\u901a\u5e38\u7531\u4e00\u4e2a\u5d4c\u5165\u6295\u5f71\u5c42\u548c\u591a\u4e2a\u76f8\u540c\u7684Transformer\u5c42\u6784\u6210\u3002\u5728\u63a8\u7406\u8fc7\u7a0b\u4e2d\uff0c\u5404\u5c42\u987a\u5e8f\u6267\u884c\uff0c\u4e0a\u4e00\u5c42\u8f93\u51fa\u4f5c\u4e3a\u4e0b\u4e00\u5c42\u8f93\u5165\u3002<\/code>AirLLM\u5229\u7528\u8fd9\u4e00\u7279\u6027\uff0c\u5728\u6267\u884c\u67d0\u4e00\u5c42\u63a8\u7406\u65f6\uff0c\u4ec5\u4ece\u78c1\u76d8\u52a0\u8f7d\u8be5\u5c42\u53c2\u6570\uff0c\u8ba1\u7b97\u5b8c\u6210\u540e\u91ca\u653e\u5185\u5b58\u3002\u4f8b\u5982\uff0c\u5bf9\u4e8e70B\u53c2\u6570\u7684\u5927\u6a21\u578b\uff0c\u6bcf\u5c42\u53ea\u9700\u7ea61.6GB GPU\u5185\u5b58\uff08\u7ea6\u4e3a\u6574\u4e2a\u6a21\u578b\u76841\/80\uff09\u3002\u5373\u4fbf\u8003\u8651\u8f93\u51fa\u7f13\u5b58\uff08\u5982KV\u7f13\u5b58\uff09\uff0c\u5f53\u8f93\u5165\u957f\u5ea6\u4e3a100\u65f6\uff0c\u989d\u5916GPU\u5185\u5b58\u4ec5\u7ea630MB\uff0c\u8fd9\u4f7f\u5f97\u6574\u4e2a70B\u6a21\u578b\u63a8\u7406\u57284GB\u663e\u5b58\u5185\u5373\u53ef\u5b8c\u6210\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">2\u3001<strong>FlashAttention<\/strong><br><code>FlashAttention\u662fAirLLM\u6027\u80fd\u63d0\u5347\u7684\u5173\u952e\u6280\u672f\u3002<\/code>\u5b83\u6e90\u4e8e\u5bf9self-attention\u5185\u5b58\u590d\u6742\u5ea6\u4f18\u5316\u7684\u7814\u7a76\uff0c\u5c06\u539f\u59cbO(n\u00b2)\u7684\u5185\u5b58\u590d\u6742\u5ea6\u964d\u4f4e\u81f3O(n)\u3002\u901a\u8fc7\u6309\u987a\u5e8f\u8ba1\u7b97\u548c\u66f4\u65b0\u4e2d\u95f4\u7ed3\u679c\uff0c\u5e76\u4e22\u5f03\u4e0d\u5fc5\u8981\u6570\u636e\u7684\u65b9\u5f0f\uff0c\u5b9e\u73b0\u4e86\u8fd9\u4e00\u7a81\u7834\u3002\u540c\u65f6\uff0cFlashAttention\u6df1\u5ea6\u4f18\u5316\u4e86CUDA\u5185\u5b58\u8bbf\u95ee\uff0c\u4ece\u800c\u5728\u63a8\u7406\u548c\u8bad\u7ec3\u9636\u6bb5\u5b9e\u73b0\u591a\u500d\u52a0\u901f\u3002\u8fd9\u79cd\u4f18\u5316\u4f7f\u5f97\u6a21\u578b\u5728\u5904\u7406\u957f\u5e8f\u5217\u6587\u672c\u65f6\u66f4\u52a0\u9ad8\u6548\uff0c\u663e\u8457\u63d0\u5347\u4e86\u63a8\u7406\u901f\u5ea6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">3\u3001<strong>\u6a21\u578b\u6587\u4ef6\u5171\u4eab<\/strong><br><code>\u539f\u59cb\u5927\u578b\u8bed\u8a00\u6a21\u578b\u6587\u4ef6\u901a\u5e38\u5206\u5272\u4e3a\u8f83\u5927\u5757\uff08\u598210GB\u5de6\u53f3\uff09\uff0c\u4f46\u5206\u5c42\u63a8\u7406\u6bcf\u5c42\u4ec5\u9700\u7ea61.6GB\u6570\u636e\u3002<\/code>\u82e5\u6309\u539f\u59cb\u5757\u52a0\u8f7d\uff0c\u6bcf\u5c42\u6267\u884c\u90fd\u8981\u91cd\u65b0\u52a0\u8f7d\u6574\u4e2a\u6587\u4ef6\uff0c\u4f1a\u9020\u6210\u5927\u91cf\u5185\u5b58\u6d6a\u8d39\u3002AirLLM\u901a\u8fc7\u9884\u5904\u7406\u548c\u5206\u5c42\u5206\u5272\u539f\u59cbHugging Face\u6a21\u578b\u6587\u4ef6\uff0c\u5e76\u4f7f\u7528\u5b89\u5168\u5f20\u91cf\uff08SafeTensors\uff09\u5b58\u50a8\uff0c\u4ee5\u5185\u5b58\u6620\u5c04\u65b9\u5f0f\u52a0\u8f7d\u3002SafeTensors\u786e\u4fdd\u5b58\u50a8\u548c\u5185\u5b58\u683c\u5f0f\u5339\u914d\uff0c\u52a0\u5feb\u78c1\u76d8\u8bfb\u53d6\u901f\u5ea6\uff0c\u51cf\u5c11\u91cd\u590d\u52a0\u8f7d\u548c\u78c1\u76d8\u8bfb\u53d6\u5bfc\u81f4\u7684\u5185\u5b58\u635f\u8017\uff0c\u6709\u6548\u63d0\u9ad8\u4e86\u63a8\u7406\u6548\u7387\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">4\u3001<strong>\u5143\u8bbe\u5907<\/strong><br><code>AirLLM\u501f\u52a9Hugging Face Accelerate\u7684\u5143\u8bbe\u5907\uff08MetaDevice\uff09\u529f\u80fd\u4f18\u5316\u5185\u5b58\u3002<\/code>\u5143\u8bbe\u5907\u662f\u4e3a\u8fd0\u884c\u8d85\u5927\u578b\u6a21\u578b\u8bbe\u8ba1\u7684\u865a\u62df\u8bbe\u5907\uff0c\u901a\u8fc7\u5143\u8bbe\u5907\u52a0\u8f7d\u6a21\u578b\u65f6\uff0c\u4ec5\u52a0\u8f7d\u4ee3\u7801\uff0c\u5185\u5b58\u4f7f\u7528\u7387\u4e3a\u96f6\u3002\u5728\u6267\u884c\u8fc7\u7a0b\u4e2d\uff0c\u6839\u636e\u9700\u8981\u5c06\u6a21\u578b\u90e8\u5206\u5185\u5bb9\u4ece\u5143\u8bbe\u5907\u52a8\u6001\u8f6c\u79fb\u5230CPU\u6216GPU\u7b49\u771f\u5b9e\u8bbe\u5907\u3002\u8fd9\u79cd\u52a8\u6001\u52a0\u8f7d\u673a\u5236\u4f7f\u6a21\u578b\u80fd\u6839\u636e\u786c\u4ef6\u8d44\u6e90\u7075\u6d3b\u8c03\u6574\uff0c\u5728\u4fdd\u8bc1\u63a8\u7406\u6027\u80fd\u7684\u540c\u65f6\uff0c\u6700\u5927\u7a0b\u5ea6\u51cf\u5c11\u5185\u5b58\u5360\u7528\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_25\"><\/a>\u4e8c\u3001\u529f\u80fd\u7279\u70b9<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">1\u3001<strong>\u4f4e\u5185\u5b58\u9700\u6c42<\/strong><br><code>AirLLM\u7684\u7a81\u51fa\u4f18\u52bf\u5728\u4e8e\u5176\u4f4e\u5185\u5b58\u9700\u6c42\u3002<\/code>\u5b83\u6253\u7834\u4e86\u5927\u578b<a href=\"https:\/\/so.csdn.net\/so\/search?q=%E8%AF%AD%E8%A8%80%E6%A8%A1%E5%9E%8B&amp;spm=1001.2101.3001.7020\" target=\"_blank\" rel=\"noreferrer noopener\">\u8bed\u8a00\u6a21\u578b<\/a>\u5bf9\u9ad8\u7aef\u786c\u4ef6\u7684\u4f9d\u8d56\uff0c\u80fd\u8ba970\u4ebf\u53c2\u6570\u7ea7\u522b\u7684\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u4ec54GB VRAM\u7684\u5355\u5f20GPU\u5361\u4e0a\u8fd0\u884c\uff0c\u4e14\u65e0\u9700\u91cf\u5316\u3001\u84b8\u998f\u6216\u526a\u679d\u7b49\u6a21\u578b\u538b\u7f29\u624b\u6bb5\u3002\u66f4\u5389\u5bb3\u7684\u662f\uff0c\u5b83\u53ef\u4f7f405B\u53c2\u6570\u7684Llama 3.1\u6a21\u578b\u57288GB VRAM\u7684\u786c\u4ef6\u4e0a\u8fd0\u884c\uff0c\u6781\u5927\u964d\u4f4e\u4e86\u5927\u6a21\u578b\u63a8\u7406\u7684\u786c\u4ef6\u95e8\u69db\uff0c\u8ba9\u66f4\u591a\u8bbe\u5907\u80fd\u8fd0\u884c\u5927\u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">2\u3001<strong>\u9ad8\u5ea6\u7075\u6d3b\u6027<\/strong><br><code>AirLLM\u9002\u7528\u4e8e\u591a\u79cd\u573a\u666f\uff0c<\/code>\u65e0\u8bba\u662f\u5b66\u672f\u7814\u7a76\u4e2d\u7684\u6a21\u578b\u63a2\u7d22\u3001\u5de5\u4e1a\u751f\u4ea7\u4e2d\u7684\u5e94\u7528\u5f00\u53d1\uff0c\u8fd8\u662f\u5728\u79fb\u52a8\u8bbe\u5907\u3001\u8d44\u6e90\u53d7\u9650\u7684\u6570\u636e\u4e2d\u5fc3\u90e8\u7f72AI\u5e94\u7528\uff0c\u90fd\u80fd\u8868\u73b0\u51fa\u8272\u3002\u5b83\u80fd\u6839\u636e\u4e0d\u540c\u786c\u4ef6\u6761\u4ef6\u548c\u5e94\u7528\u573a\u666f\u81ea\u9002\u5e94\u8c03\u6574\u6a21\u578b\u8fd0\u884c\u65b9\u5f0f\uff0c\u6ee1\u8db3\u591a\u6837\u5316\u9700\u6c42\uff0c\u63a8\u52a8AI\u5728\u5404\u9886\u57df\u5e7f\u6cdb\u5e94\u7528\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">3\u3001<strong>\u5f00\u6e90\u7279\u6027<\/strong><br>\u4f5c\u4e3a\u5f00\u6e90\u9879\u76ee\uff0cAirLLM\u5438\u5f15\u4e86\u5168\u7403\u5f00\u53d1\u8005\u53c2\u4e0e\u3002\u8fd9\u79cd\u5f00\u6e90\u6a21\u5f0f\u4fc3\u8fdb\u4e86\u6280\u672f\u4ea4\u6d41\u4e0e\u534f\u4f5c\uff0c\u5f00\u53d1\u8005\u53ef\u8d21\u732e\u4ee3\u7801\u3001\u63d0\u51fa\u6539\u8fdb\u5efa\u8bae\uff0c\u5171\u540c\u63a8\u52a8\u9879\u76ee\u53d1\u5c55\u3002\u540c\u65f6\uff0c\u5f00\u6e90\u4e5f\u4fbf\u4e8e\u5f00\u53d1\u8005\u6839\u636e\u81ea\u8eab\u9700\u6c42\u5b9a\u5236\u548c\u6269\u5c55\u529f\u80fd\uff0c\u52a0\u901f\u4e86\u5927\u578b\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u6280\u672f\u7684\u521b\u65b0\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_35\"><\/a>\u4e09\u3001\u5e94\u7528\u573a\u666f<\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<ul class=\"wp-block-list\">\n<li>1\u3001<strong>\u6559\u80b2\u79d1\u7814\u9886\u57df<\/strong><br>\u5728\u6559\u80b2\u548c\u79d1\u7814\u4e2d\uff0cAirLLM\u610f\u4e49\u91cd\u5927\u3002\u8fc7\u53bb\uff0c\u63a2\u7d22\u5148\u8fdbNLP\u6a21\u578b\u9700\u8981\u6602\u8d35\u786c\u4ef6\uff0c\u9650\u5236\u4e86\u9ad8\u6821\u5b9e\u9a8c\u5ba4\u548c\u4e2a\u4eba\u7814\u7a76\u8005\u3002\u73b0\u5728\uff0c\u501f\u52a9AirLLM\uff0c\u4ed6\u4eec\u5728\u672c\u5730\u666e\u901a\u8bbe\u5907\u4e0a\u5c31\u80fd\u8fd0\u884c70B\u751a\u81f3\u66f4\u5927\u53c2\u6570\u7684\u6a21\u578b\uff0c\u964d\u4f4e\u79d1\u7814\u95e8\u69db\uff0c\u6fc0\u53d1\u79d1\u7814\u521b\u65b0\uff0c\u63a8\u52a8<a href=\"https:\/\/so.csdn.net\/so\/search?q=%E8%87%AA%E7%84%B6%E8%AF%AD%E8%A8%80%E5%A4%84%E7%90%86&amp;spm=1001.2101.3001.7020\" target=\"_blank\" rel=\"noreferrer noopener\">\u81ea\u7136\u8bed\u8a00\u5904\u7406<\/a>\u7814\u7a76\u53d1\u5c55\uff0c\u57f9\u517b\u66f4\u591aAI\u4eba\u624d\u3002<\/li>\n\n\n\n<li>2\u3001<strong>\u4f01\u4e1a\u7ea7\u5e94\u7528\u5f00\u53d1<\/strong><br>\u5bf9\u4e8e\u4f01\u4e1a\u5f00\u53d1\u56e2\u961f\uff0cAirLLM\u52a9\u529b\u5feb\u901f\u5b9e\u73b0AI\u529f\u80fd\u3002\u5728\u6784\u5efa\u667a\u80fd\u5ba2\u670d\u3001\u5185\u5bb9\u751f\u6210\u5e73\u53f0\u3001\u6570\u636e\u5206\u6790\u9884\u6d4b\u7b49\u5e94\u7528\u65f6\uff0c\u53ef\u57fa\u4e8e\u73b0\u6709\u786c\u4ef6\u8d44\u6e90\u5229\u7528AirLLM\u8fd0\u884c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff0c\u52a0\u901f\u4f01\u4e1a\u6570\u5b57\u5316\u8f6c\u578b\u548cAI\u5e94\u7528\u843d\u5730\uff0c\u63d0\u5347\u4f01\u4e1a\u7ade\u4e89\u529b\u548c\u521b\u65b0\u80fd\u529b\u3002<\/li>\n\n\n\n<li>3\u3001<strong>\u4e2a\u4eba\u5f00\u53d1\u8005\u521b\u65b0<\/strong><br>\u4e2a\u4eba\u5f00\u53d1\u8005\u4e5f\u56e0AirLLM\u53d7\u76ca\u3002\u4ed6\u4eec\u80fd\u5728\u7b14\u8bb0\u672c\u7535\u8111\u7b49\u666e\u901a\u786c\u4ef6\u4e0a\u8fd0\u884c\u5927\u89c4\u6a21\u6a21\u578b\uff0c\u8fdb\u884c\u6d4b\u8bd5\u3001\u5f00\u53d1\u548c\u521b\u65b0\u3002\u8fd9\u4e3a\u4e2a\u4eba\u5f00\u53d1\u8005\u63d0\u4f9b\u5e7f\u9614\u521b\u4f5c\u7a7a\u95f4\uff0c\u4f7f\u5176\u80fd\u81ea\u7531\u63a2\u7d22AI\u5e94\u7528\u8fb9\u754c\uff0c\u5f00\u53d1\u5982\u5199\u4f5c\u52a9\u624b\u3001\u667a\u80fd\u5b66\u4e60\u5de5\u5177\u7b49\u521b\u65b0\u5e94\u7528\u3002<\/li>\n<\/ul>\n<\/blockquote>\n\n\n\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_46\"><\/a>\u56db\u3001\u4f7f\u7528\u65b9\u6cd5<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">1\u3001\u5b89\u88c5 airllm \u5305<br>\u901a\u8fc7<code>pip install airllm<\/code>\u547d\u4ee4\u5b89\u88c5AirLLM\u5305\u3002<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>pip install airllm\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">2\u3001\u6a21\u578b\u63a8\u7406\u793a\u4f8b\uff08\u4ee5Llama 2\u6a21\u578b\u4e3a\u4f8b\uff09<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>from airllm import AutoModel\n\n# \u5b9a\u4e49\u6a21\u578b\u8f93\u5165\u7684\u6700\u5927\u957f\u5ea6\nMAX_LENGTH = 128\n\n# \u4f7f\u7528\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u8def\u5f84\u6216\u540d\u79f0\u6765\u52a0\u8f7d\u6a21\u578b\n# \u8fd9\u91cc\u4f7f\u7528\u7684\u662fHugging Face\u6a21\u578b\u5e93\u4e2d\u7684\u6a21\u578bID\nmodel = AutoModel.from_pretrained(\"garage-bAInd\/Platypus2-70B-instruct\")\n\n# \u5982\u679c\u6a21\u578b\u5df2\u7ecf\u4e0b\u8f7d\u5230\u672c\u5730\uff0c\u4e5f\u53ef\u4ee5\u4f7f\u7528\u672c\u5730\u8def\u5f84\u6765\u52a0\u8f7d\u6a21\u578b\n# \u8fd9\u4e00\u884c\u88ab\u6ce8\u91ca\u6389\u4e86\uff0c\u5982\u679c\u9700\u8981\u4f7f\u7528\u672c\u5730\u6a21\u578b\uff0c\u53ef\u4ee5\u53d6\u6d88\u6ce8\u91ca\n#model = AutoModel.from_pretrained(\"\/home\/ubuntu\/.cache\/huggingface\/hub\/models--garage-bAInd--Platypus2-70B-instruct\/snapshots\/b585e74bcaae02e52665d9ac6d23f4d0dbc81a0f\")\n\n# \u5b9a\u4e49\u8f93\u5165\u6587\u672c\u5217\u8868\uff0c\u8fd9\u91cc\u53ea\u6709\u4e00\u4e2a\u95ee\u9898\ninput_text = &#91;\n    'What is the capital of United States?',\n    #'I like',\n]\n\n# \u4f7f\u7528\u6a21\u578b\u7684tokenizer\u5c06\u8f93\u5165\u6587\u672c\u8f6c\u6362\u4e3a\u6a21\u578b\u53ef\u4ee5\u7406\u89e3\u7684token\n# return_tensors=\"pt\" \u8868\u793a\u8fd4\u56dePyTorch\u5f20\u91cf\n# return_attention_mask=False \u8868\u793a\u4e0d\u8fd4\u56de\u6ce8\u610f\u529b\u63a9\u7801\n# truncation=True \u8868\u793a\u5982\u679c\u8f93\u5165\u8d85\u8fc7\u6700\u5927\u957f\u5ea6\u5219\u622a\u65ad\n# max_length=MAX_LENGTH \u8bbe\u7f6e\u6700\u5927\u957f\u5ea6\u9650\u5236\n# padding=False \u8868\u793a\u4e0d\u8fdb\u884c\u586b\u5145\ninput_tokens = model.tokenizer(input_text,\n    return_tensors=\"pt\", \n    return_attention_mask=False, \n    truncation=True, \n    max_length=MAX_LENGTH, \n    padding=False)\n\n# \u4f7f\u7528\u6a21\u578b\u751f\u6210\u6587\u672c\n# input_tokens&#91;'input_ids'].cuda() \u5c06\u8f93\u5165ID\u4f20\u9012\u5230GPU\n# max_new_tokens=20 \u8bbe\u7f6e\u751f\u6210\u7684\u6700\u5927\u65b0token\u6570\u91cf\n# use_cache=True \u8868\u793a\u4f7f\u7528\u7f13\u5b58\u6765\u52a0\u901f\u751f\u6210\u8fc7\u7a0b\n# return_dict_in_generate=True \u8868\u793a\u8fd4\u56de\u4e00\u4e2a\u5b57\u5178\u800c\u4e0d\u662f\u76f4\u63a5\u8fd4\u56de\u751f\u6210\u7684\u5e8f\u5217\ngeneration_output = model.generate(\n    input_tokens&#91;'input_ids'].cuda(), \n    max_new_tokens=20,\n    use_cache=True,\n    return_dict_in_generate=True)\n\n# \u5c06\u751f\u6210\u7684token\u5e8f\u5217\u89e3\u7801\u56de\u6587\u672c\noutput = model.tokenizer.decode(generation_output.sequences&#91;0])\n\n# \u6253\u5370\u751f\u6210\u7684\u6587\u672c\nprint(output)\n<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\"><a><\/a><a id=\"_105\"><\/a>\u7ed3\u8bed<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">AirLLM\u4f5c\u4e3a\u5927\u578b\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u9886\u57df\u7684\u521b\u65b0\u5f00\u6e90\u5de5\u5177\u5305\uff0c\u51ed\u501f\u5176\u5353\u8d8a\u7684\u6280\u672f\u4eae\u70b9\u3001\u7a81\u51fa\u7684\u529f\u80fd\u7279\u70b9\u548c\u5e7f\u6cdb\u7684\u5e94\u7528\u573a\u666f\uff0c\u4e3a\u89e3\u51b3\u5927\u6a21\u578b\u5728\u8d44\u6e90\u53d7\u9650\u73af\u5883\u4e0b\u7684\u63a8\u7406\u96be\u9898\u63d0\u4f9b\u4e86\u6709\u6548\u65b9\u6848\u3002\u5b83\u63a8\u52a8\u4e86AI\u6280\u672f\u7684\u666e\u53ca\u548c\u53d1\u5c55\uff0c\u8ba9\u66f4\u591a\u4eba\u80fd\u591f\u5229\u7528\u5927\u578b\u8bed\u8a00\u6a21\u578b\u521b\u9020\u4ef7\u503c\u3002\u65e0\u8bba\u662f\u79d1\u7814\u4eba\u5458\u3001\u4f01\u4e1a\u5f00\u53d1\u8005\u8fd8\u662f\u4e2a\u4eba\u5f00\u53d1\u8005\uff0c\u90fd\u80fd\u4eceAirLLM\u4e2d\u83b7\u5f97\u52a9\u529b\uff0c\u5f00\u542f\u66f4\u591a\u7684\u53ef\u80fd\u6027\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GitHub\u9879\u76ee\u5730\u5740\uff1a<a href=\"https:\/\/github.com\/lyogavin\/airllm\">https:\/\/github.com\/lyogavin\/airllm<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u539f\u6587\u94fe\u63a5\uff1ahttps:\/\/blog.csdn.net\/xiaobing259\/article\/details\/143836815<br>\u4f5c\u8005\uff1a\u5bfb\u9053AI\u5c0f\u5175<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u524d\u8a00 \u5728\u4eba\u5de5\u667a\u80fd\u84ec\u52c3\u53d1\u5c55\u7684\u6d6a\u6f6e\u4e2d\uff0c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09&#8230; <\/p>\n","protected":false},"author":1,"featured_media":485,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[121,122,123],"tags":[],"class_list":["post-483","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-122","category-123"],"_links":{"self":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/483","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=483"}],"version-history":[{"count":1,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/483\/revisions"}],"predecessor-version":[{"id":484,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/483\/revisions\/484"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/media\/485"}],"wp:attachment":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=483"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=483"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=483"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}