{"id":55,"date":"2024-07-21T15:39:32","date_gmt":"2024-07-21T07:39:32","guid":{"rendered":"https:\/\/www.onetts.com\/ai\/?p=55"},"modified":"2024-07-21T18:14:34","modified_gmt":"2024-07-21T10:14:34","slug":"qwen1-5-moe-a2-7b","status":"publish","type":"post","link":"https:\/\/www.onetts.com\/ai\/models\/qwen1-5-moe-a2-7b\/","title":{"rendered":"Qwen1.5-MoE-A2.7B"},"content":{"rendered":"<p data-spm-anchor-id=\"a2c6h.13066512.0.i3.763536afMBFw2G\">Qwen1.5-MoE-A2.7B\u662fQwen\u7cfb\u5217\u63a8\u51fa\u7684\u9996\u4e2aMixture of Experts\uff08MoE\uff09\u6a21\u578b\u3002\u5b83\u4ee5\u5176\u8f83\u5c0f\u7684\u6fc0\u6d3b\u53c2\u6570\u91cf\uff0827\u4ebf\uff09\u5b9e\u73b0\u4e86\u4e0e70\u4ebf\u53c2\u6570\u6a21\u578b\u76f8\u5f53\u7684\u6027\u80fd\uff0c\u663e\u8457\u63d0\u5347\u4e86\u6548\u7387\u548c\u53ef\u6269\u5c55\u6027\u3002\u8be5\u6a21\u578b\u5728\u8bbe\u8ba1\u4e0a\u91c7\u7528\u4e86\u7279\u522b\u4f18\u5316\u7684MoE\u67b6\u6784\uff0c\u901a\u8fc7\u7cbe\u7ec6\u7684\u4e13\u5bb6\uff08expert\uff09\u8bbe\u8ba1\u548c\u521b\u65b0\u7684\u8def\u7531\u673a\u5236\uff0c\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u53c2\u6570\u5229\u7528\u548c\u5feb\u901f\u7684\u63a8\u7406\u901f\u5ea6\u3002<\/p>\n<h2 id=\"-\">\u6a21\u578b\u8bc4\u6d4b<\/h2>\n<p>Qwen1.5-MoE-A2.7B\u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u5c55\u73b0\u4e86\u5353\u8d8a\u7684\u6027\u80fd\u3002\u4ee5\u4e0b\u662f\u4e00\u4e9b\u5173\u952e\u8bc4\u6d4b\u7ed3\u679c\uff1a<\/p>\n<ul>\n<li><strong>MMLU<\/strong>\uff1a\u5728\u591a\u8bed\u8a00\u7406\u89e3\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cQwen1.5-MoE-A2.7B\u5f97\u5206\u4e3a62.5\uff0c\u63a5\u8fd1\u9876\u5c16\u76847B\u6a21\u578b\u3002<\/li>\n<li><strong>GSM8K<\/strong>\uff1a\u5728\u901a\u7528\u79d1\u5b66\u77e5\u8bc6\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5f97\u5206\u4e3a61.5\uff0c\u663e\u793a\u51fa\u5176\u5728\u79d1\u5b66\u77e5\u8bc6\u7406\u89e3\u65b9\u9762\u7684\u5f3a\u5927\u80fd\u529b\u3002<\/li>\n<li><strong>HumanEval<\/strong>\uff1a\u5728\u4eba\u7c7b\u8bc4\u4f30\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5f97\u5206\u4e3a34.2\uff0c\u4f53\u73b0\u4e86\u5176\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u7684\u51fa\u8272\u8868\u73b0\u3002<\/li>\n<li><strong>Multilingual<\/strong>\uff1a\u5728\u591a\u8bed\u8a00\u7efc\u5408\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5f97\u5206\u4e3a40.8\uff0c\u8bc1\u660e\u4e86\u5176\u5728\u591a\u8bed\u8a00\u5904\u7406\u65b9\u9762\u7684\u4f18\u52bf\u3002<\/li>\n<li><strong>MT-Bench<\/strong>\uff1a\u5728\u804a\u5929\u6a21\u578b\u6d4b\u8bd5\u4e2d\uff0c\u5f97\u5206\u4e3a7.17\uff0c\u663e\u793a\u51fa\u5176\u5728\u5bf9\u8bdd\u751f\u6210\u65b9\u9762\u7684\u6f5c\u529b\u3002<\/li>\n<\/ul>\n<h2 id=\"-\">\u90e8\u7f72\u4f7f\u7528<\/h2>\n<p>Qwen1.5-MoE-A2.7B\u6a21\u578b\u7684\u90e8\u7f72\u548c\u4f7f\u7528\u76f8\u5bf9\u7b80\u5355\uff0c\u4e3b\u8981\u6b65\u9aa4\u5982\u4e0b\uff1a<\/p>\n<ol>\n<li><strong>\u5b89\u88c5<\/strong>\uff1a\u9700\u8981\u4ece\u6e90\u4ee3\u7801\u5b89\u88c5transformers\u548cvLLM\u5e93\u3002\n<pre><code class=\"lang-bash\"><span class=\"token function\">git<\/span> clone https:\/\/github.com\/huggingface\/transformers\r\n<span class=\"token function\">cd<\/span> transformers\r\npip <span class=\"token function\">install<\/span> -e <span class=\"token keyword\">.<\/span>\r\n<\/code><\/pre>\n<pre><code class=\"lang-bash\"><span class=\"token function\">git<\/span> clone https:\/\/github.com\/vllm-project\/vllm.git\r\n<span class=\"token function\">cd<\/span> vllm\r\npip <span class=\"token function\">install<\/span> -e <span class=\"token keyword\">.<\/span>\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u6a21\u578b\u52a0\u8f7d<\/strong>\uff1a\u4f7f\u7528transformers\u5e93\u52a0\u8f7d\u6a21\u578b\u3002\n<pre><code class=\"lang-python\"><span class=\"token keyword\">from<\/span> transformers <span class=\"token keyword\">import<\/span> AutoModelForCausalLM<span class=\"token punctuation\">,<\/span> AutoTokenizer\r\n\r\nmodel <span class=\"token operator\">=<\/span> AutoModelForCausalLM<span class=\"token punctuation\">.<\/span>from_pretrained<span class=\"token punctuation\">(<\/span>\r\n    <span class=\"token string\">\"Qwen\/Qwen1.5-MoE-A2.7B-Chat\"<\/span><span class=\"token punctuation\">,<\/span>\r\n    device_map<span class=\"token operator\">=<\/span><span class=\"token string\">\"auto\"<\/span>\r\n<span class=\"token punctuation\">)<\/span>\r\ntokenizer <span class=\"token operator\">=<\/span> AutoTokenizer<span class=\"token punctuation\">.<\/span>from_pretrained<span class=\"token punctuation\">(<\/span><span class=\"token string\">\"Qwen\/Qwen1.5-MoE-A2.7B-Chat\"<\/span><span class=\"token punctuation\">)<\/span>\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u751f\u6210\u6587\u672c<\/strong>\uff1a\u901a\u8fc7\u6a21\u578b\u751f\u6210\u6587\u672c\u3002\n<pre><code class=\"lang-python\">prompt <span class=\"token operator\">=<\/span> <span class=\"token string\">\"Give me a short introduction to large language model.\"<\/span>\r\nmessages <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span>\r\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">\"role\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token string\">\"system\"<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\"content\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token string\">\"You are a helpful assistant.\"<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span>\r\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">\"role\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token string\">\"user\"<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\"content\"<\/span><span class=\"token punctuation\">:<\/span> prompt<span class=\"token punctuation\">}<\/span>\r\n<span class=\"token punctuation\">]<\/span>\r\ntext <span class=\"token operator\">=<\/span> tokenizer<span class=\"token punctuation\">.<\/span>apply_chat_template<span class=\"token punctuation\">(<\/span>\r\n    messages<span class=\"token punctuation\">,<\/span>\r\n    tokenize<span class=\"token operator\">=<\/span><span class=\"token boolean\">False<\/span><span class=\"token punctuation\">,<\/span>\r\n    add_generation_prompt<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span>\r\n<span class=\"token punctuation\">)<\/span>\r\nmodel_inputs <span class=\"token operator\">=<\/span> tokenizer<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">[<\/span>text<span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span> return_tensors<span class=\"token operator\">=<\/span><span class=\"token string\">\"pt\"<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">.<\/span>to<span class=\"token punctuation\">(<\/span>model<span class=\"token punctuation\">.<\/span>device<span class=\"token punctuation\">)<\/span>\r\n\r\ngenerated_ids <span class=\"token operator\">=<\/span> model<span class=\"token punctuation\">.<\/span>generate<span class=\"token punctuation\">(<\/span>\r\n    model_inputs<span class=\"token punctuation\">.<\/span>input_ids<span class=\"token punctuation\">,<\/span>\r\n    max_new_tokens<span class=\"token operator\">=<\/span><span class=\"token number\">512<\/span>\r\n<span class=\"token punctuation\">)<\/span>\r\nresponse <span class=\"token operator\">=<\/span> tokenizer<span class=\"token punctuation\">.<\/span>batch_decode<span class=\"token punctuation\">(<\/span>generated_ids<span class=\"token punctuation\">,<\/span> skip_special_tokens<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span>\r\n<\/code><\/pre>\n<\/li>\n<li><strong>API\u63a5\u53e3<\/strong>\uff1a\u4f7f\u7528vLLM\u6784\u5efa\u4e0e\u6a21\u578b\u517c\u5bb9\u7684OpenAI-API\u63a5\u53e3\u3002\n<pre><code class=\"lang-bash\">python -m vllm.entrypoints.openai.api_server --model Qwen\/Qwen1.5-MoE-A2.7B-Chat\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h2 id=\"-\">\u5e38\u89c1\u95ee\u9898<\/h2>\n<ol>\n<li><strong>\u6a21\u578b\u517c\u5bb9\u6027<\/strong>\uff1aQwen1.5-MoE-A2.7B\u662f\u5426\u4e0e\u5176\u4ed6\u6846\u67b6\u517c\u5bb9\uff1f\n<ul>\n<li>\u76ee\u524d\uff0c\u6a21\u578b\u4e3b\u8981\u901a\u8fc7transformers\u548cvLLM\u5e93\u8fdb\u884c\u90e8\u7f72\uff0c\u672a\u6765\u53ef\u80fd\u4f1a\u6269\u5c55\u5230\u66f4\u591a\u6846\u67b6\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u63a8\u7406\u901f\u5ea6<\/strong>\uff1aQwen1.5-MoE-A2.7B\u7684\u63a8\u7406\u901f\u5ea6\u5982\u4f55\uff1f\n<ul>\n<li>\u76f8\u6bd4Qwen1.5-7B\uff0c\u63a8\u7406\u901f\u5ea6\u63d0\u5347\u4e86\u7ea61.74\u500d\uff0c\u4e3b\u8981\u5f97\u76ca\u4e8eMoE\u6a21\u578b\u7684\u7a00\u758f\u6fc0\u6d3b\u7279\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u8bad\u7ec3\u6210\u672c<\/strong>\uff1aQwen1.5-MoE-A2.7B\u7684\u8bad\u7ec3\u6210\u672c\u5982\u4f55\uff1f\n<ul>\n<li>\u8bad\u7ec3\u6210\u672c\u964d\u4f4e\u4e8675%\uff0c\u4e3b\u8981\u7531\u4e8e\u5176\u8f83\u5c11\u7684\u6fc0\u6d3b\u53c2\u6570\u548c\u9ad8\u6548\u7684\u521d\u59cb\u5316\u65b9\u6cd5\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h2 id=\"-\">\u76f8\u5173\u8d44\u6e90<\/h2>\n<ul>\n<li><strong>\u6a21\u578b\u4ee3\u7801<\/strong>\uff1a<a target=\"_blank\" rel=\"noopener noreferrer nofollow\" href=\"https:\/\/github.com\/qwenlm\/qwen-moe\">Qwen1.5-MoE-A2.7B GitHub<\/a><\/li>\n<li><strong>\u6587\u6863<\/strong>\uff1a<a target=\"_blank\" rel=\"noopener noreferrer nofollow\" href=\"https:\/\/qwenlm.github.io\/zh\/blog\/qwen-moe\/\">Qwen1.5-MoE-A2.7B \u6587\u6863<\/a><\/li>\n<li><strong>\u793e\u533a\u652f\u6301<\/strong>\uff1a\u53ef\u4ee5\u901a\u8fc7Hugging Face\u548cvLLM\u793e\u533a\u83b7\u53d6\u66f4\u591a\u652f\u6301\u548c\u8d44\u6e90\u3002<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u8fd9\u4e9b\u8d44\u6e90\uff0c\u7528\u6237\u53ef\u4ee5\u66f4\u6df1\u5165\u5730\u4e86\u89e3Qwen1.5-MoE-A2.7B\u6a21\u578b\uff0c\u5e76\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u5145\u5206\u5229\u7528\u5176\u4f18\u52bf\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Qwen1.5-MoE-A2.7B\u662fQwen\u7cfb\u5217\u63a8\u51fa\u7684\u9996\u4e2aMixture of Experts\uff08MoE\uff09\u6a21\u578b\u3002\u5b83\u4ee5\u5176\u8f83\u5c0f\u7684\u6fc0\u6d3b\u53c2\u6570\u91cf\uff0827\u4ebf\uff09\u5b9e\u73b0\u4e86\u4e0e70\u4ebf\u53c2\u6570\u6a21\u578b\u76f8\u5f53\u7684\u6027\u80fd\uff0c\u663e\u8457\u63d0\u5347\u4e86\u6548\u7387\u548c\u53ef\u6269\u5c55\u6027\u3002\u8be5\u6a21\u578b\u5728\u8bbe\u8ba1\u4e0a\u91c7\u7528\u4e86\u7279\u522b\u4f18\u5316\u7684MoE\u67b6\u6784\uff0c\u901a\u8fc7\u7cbe\u7ec6\u7684\u4e13\u5bb6\uff08expert\uff09\u8bbe\u8ba1\u548c\u521b\u65b0\u7684\u8def\u7531\u673a\u5236\uff0c\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u53c2\u6570\u5229\u7528\u548c\u5feb\u901f\u7684\u63a8\u7406\u901f\u5ea6\u3002 \u6a21\u578b\u8bc4\u6d4b Qwen1.5-MoE-A2.7B\u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u5c55\u73b0\u4e86\u5353\u8d8a\u7684\u6027\u80fd\u3002\u4ee5\u4e0b\u662f\u4e00\u4e9b\u5173\u952e\u8bc4\u6d4b\u7ed3\u679c\uff1a MMLU\uff1a\u5728\u591a\u8bed\u8a00\u7406\u89e3\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cQwen1.5-MoE-A2.7B\u5f97\u5206\u4e3a62.<\/p>\n","protected":false},"author":1,"featured_media":56,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[43],"collection":[42],"company":[6],"rank":[],"class_list":["post-55","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-llm","tag-moe-llm","collection-qwen1-5","company-alibaba"],"_links":{"self":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/55","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/comments?post=55"}],"version-history":[{"count":1,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/55\/revisions"}],"predecessor-version":[{"id":57,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/55\/revisions\/57"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/media\/56"}],"wp:attachment":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/media?parent=55"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/categories?post=55"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/tags?post=55"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/collection?post=55"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/company?post=55"},{"taxonomy":"rank","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/rank?post=55"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}