{"id":177,"date":"2024-07-25T16:22:03","date_gmt":"2024-07-25T08:22:03","guid":{"rendered":"https:\/\/www.onetts.com\/ai\/?p=177"},"modified":"2024-07-25T16:55:50","modified_gmt":"2024-07-25T08:55:50","slug":"glm-4-9b-chat","status":"publish","type":"post","link":"https:\/\/www.onetts.com\/ai\/models\/glm-4-9b-chat\/","title":{"rendered":"GLM-4-9B-Chat"},"content":{"rendered":"<p data-spm-anchor-id=\"a2c6h.13066512.0.i3.3cf036afMpTXID\">GLM-4-9B-Chat\u662f\u667a\u8c31AI\u63a8\u51fa\u7684\u6700\u65b0\u4e00\u4ee3\u9884\u8bad\u7ec3\u6a21\u578bGLM-4\u7cfb\u5217\u4e2d\u7684\u5f00\u6e90\u7248\u672c\u3002\u8be5\u6a21\u578b\u5728\u8bed\u4e49\u7406\u89e3\u3001\u6570\u5b66\u63a8\u7406\u3001\u4ee3\u7801\u6267\u884c\u548c\u77e5\u8bc6\u83b7\u53d6\u7b49\u591a\u4e2a\u65b9\u9762\u8868\u73b0\u51fa\u5353\u8d8a\u7684\u6027\u80fd\u3002\u5b83\u4e0d\u4ec5\u80fd\u591f\u8fdb\u884c\u591a\u8f6e\u5bf9\u8bdd\uff0c\u8fd8\u5177\u5907\u7f51\u9875\u6d4f\u89c8\u3001\u4ee3\u7801\u6267\u884c\u3001\u81ea\u5b9a\u4e49\u5de5\u5177\u8c03\u7528\u548c\u957f\u6587\u672c\u63a8\u7406\u7b49\u9ad8\u7ea7\u529f\u80fd\u3002\u6a21\u578b\u652f\u6301\u5305\u62ec\u65e5\u8bed\u3001\u97e9\u8bed\u3001\u5fb7\u8bed\u5728\u5185\u768426\u79cd\u8bed\u8a00\uff0c\u8fdb\u4e00\u6b65\u589e\u5f3a\u4e86\u5176\u591a\u8bed\u8a00\u5904\u7406\u80fd\u529b\u3002<\/p>\n<h2 id=\"-\">\u6a21\u578b\u8bc4\u6d4b<\/h2>\n<p>GLM-4-9B Chat\u5728\u591a\u4e2a\u7ecf\u5178\u4efb\u52a1\u4e0a\u8fdb\u884c\u4e86\u8bc4\u6d4b\uff0c\u7ed3\u679c\u663e\u793a\u5176\u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0a\u5747\u6709\u51fa\u8272\u7684\u8868\u73b0\u3002\u4ee5\u4e0b\u662f\u4e00\u4e9b\u5173\u952e\u8bc4\u6d4b\u7ed3\u679c\uff1a<\/p>\n<ul>\n<li><strong>AlignBench-v2<\/strong>: 6.61<\/li>\n<li><strong>MT-Bench<\/strong>: 8.35<\/li>\n<li><strong>IFEval<\/strong>: 69.0<\/li>\n<li><strong>MMLU<\/strong>: 72.4<\/li>\n<li><strong>C-Eval<\/strong>: 75.6<\/li>\n<li><strong>GSM8K<\/strong>: 79.6<\/li>\n<li><strong>MATH<\/strong>: 50.6<\/li>\n<li><strong>HumanEval<\/strong>: 71.8<\/li>\n<li><strong>NCB<\/strong>: 32.2<\/li>\n<\/ul>\n<p>\u6b64\u5916\uff0cGLM-4-9B-Chat\u5728\u957f\u6587\u672c\u5904\u7406\u80fd\u529b\u4e0a\u4e5f\u8fdb\u884c\u4e86\u6d4b\u8bd5\uff0c\u7ed3\u679c\u663e\u793a\u5176\u57281M\u4e0a\u4e0b\u6587\u957f\u5ea6\u4e0b\u7684\u8868\u73b0\u4f18\u5f02\u3002\u5728\u591a\u8bed\u8a00\u80fd\u529b\u6d4b\u8bd5\u4e2d\uff0cGLM-4-9B Chat\u5728\u516d\u4e2a\u591a\u8bed\u8a00\u6570\u636e\u96c6\u4e0a\u7684\u8868\u73b0\u5747\u4f18\u4e8eLlama-3-8B-Instruct\u3002<\/p>\n<h2 id=\"-\">\u90e8\u7f72\u4f7f\u7528<\/h2>\n<h4 id=\"-\">\u8be6\u7ec6\u6b65\u9aa4<\/h4>\n<ol>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong>\uff1a\n<ul>\n<li>\u786e\u4fdd\u5b89\u88c5\u4e86Python\u73af\u5883\u3002<\/li>\n<li>\u5b89\u88c5<code>transformers<\/code>\u5e93\u548c<code>torch<\/code>\u5e93\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u4e0b\u8f7d\u6a21\u578b<\/strong>\uff1a\n<pre><code class=\"lang-python\"><span class=\"token keyword\">import<\/span> torch\r\n<span class=\"token keyword\">from<\/span> transformers <span class=\"token keyword\">import<\/span> AutoModelForCausalLM<span class=\"token punctuation\">,<\/span> AutoTokenizer\r\n\r\ndevice <span class=\"token operator\">=<\/span> <span class=\"token string\">\"cuda\"<\/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\">\"THUDM\/glm-4-9b-chat\"<\/span><span class=\"token punctuation\">,<\/span> trust_remote_code<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">)<\/span>\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\">\"THUDM\/glm-4-9b-chat\"<\/span><span class=\"token punctuation\">,<\/span>\r\n    torch_dtype<span class=\"token operator\">=<\/span>torch<span class=\"token punctuation\">.<\/span>bfloat16<span class=\"token punctuation\">,<\/span>\r\n    low_cpu_mem_usage<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">,<\/span>\r\n    trust_remote_code<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span>\r\n<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">.<\/span>to<span class=\"token punctuation\">(<\/span>device<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">.<\/span>eval<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span>\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u751f\u6210\u6587\u672c<\/strong>\uff1a\n<pre><code class=\"lang-python\">query <span class=\"token operator\">=<\/span> <span class=\"token string\">\"\u4f60\u597d\"<\/span>\r\ninputs <span class=\"token operator\">=<\/span> tokenizer<span class=\"token punctuation\">.<\/span>apply_chat_template<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">[<\/span><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> query<span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span>\r\n                                      add_generation_prompt<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">,<\/span>\r\n                                      tokenize<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">,<\/span>\r\n                                      return_tensors<span class=\"token operator\">=<\/span><span class=\"token string\">\"pt\"<\/span><span class=\"token punctuation\">,<\/span>\r\n                                      return_dict<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span>\r\n                                      <span class=\"token punctuation\">)<\/span>\r\ninputs <span class=\"token operator\">=<\/span> inputs<span class=\"token punctuation\">.<\/span>to<span class=\"token punctuation\">(<\/span>device<span class=\"token punctuation\">)<\/span>\r\ngen_kwargs <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">{<\/span><span class=\"token string\">\"max_length\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">2500<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\"do_sample\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token boolean\">True<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">\"top_k\"<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1<\/span><span class=\"token punctuation\">}<\/span>\r\n<span class=\"token keyword\">with<\/span> torch<span class=\"token punctuation\">.<\/span>no_grad<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">:<\/span>\r\n    outputs <span class=\"token operator\">=<\/span> model<span class=\"token punctuation\">.<\/span>generate<span class=\"token punctuation\">(<\/span><span class=\"token operator\">**<\/span>inputs<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">**<\/span>gen_kwargs<span class=\"token punctuation\">)<\/span>\r\n    outputs <span class=\"token operator\">=<\/span> outputs<span class=\"token punctuation\">[<\/span><span class=\"token punctuation\">:<\/span><span class=\"token punctuation\">,<\/span> inputs<span class=\"token punctuation\">[<\/span><span class=\"token string\">'input_ids'<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>shape<span class=\"token punctuation\">[<\/span><span class=\"token number\">1<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">:<\/span><span class=\"token punctuation\">]<\/span>\r\n    <span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span>tokenizer<span class=\"token punctuation\">.<\/span>decode<span class=\"token punctuation\">(<\/span>outputs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><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>\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u4f7f\u7528vLLM\u540e\u7aef<\/strong>\uff1a\n<pre><code class=\"lang-python\"><span class=\"token keyword\">from<\/span> transformers <span class=\"token keyword\">import<\/span> AutoTokenizer\r\n<span class=\"token keyword\">from<\/span> vllm <span class=\"token keyword\">import<\/span> LLM<span class=\"token punctuation\">,<\/span> SamplingParams\r\n\r\nmax_model_len<span class=\"token punctuation\">,<\/span> tp_size <span class=\"token operator\">=<\/span> <span class=\"token number\">131072<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">1<\/span>\r\nmodel_name <span class=\"token operator\">=<\/span> <span class=\"token string\">\"THUDM\/glm-4-9b-chat\"<\/span>\r\nprompt <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span><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> <span class=\"token string\">\"\u4f60\u597d\"<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">]<\/span>\r\n\r\ntokenizer <span class=\"token operator\">=<\/span> AutoTokenizer<span class=\"token punctuation\">.<\/span>from_pretrained<span class=\"token punctuation\">(<\/span>model_name<span class=\"token punctuation\">,<\/span> trust_remote_code<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">)<\/span>\r\nllm <span class=\"token operator\">=<\/span> LLM<span class=\"token punctuation\">(<\/span>\r\n    model<span class=\"token operator\">=<\/span>model_name<span class=\"token punctuation\">,<\/span>\r\n    tensor_parallel_size<span class=\"token operator\">=<\/span>tp_size<span class=\"token punctuation\">,<\/span>\r\n    max_model_len<span class=\"token operator\">=<\/span>max_model_len<span class=\"token punctuation\">,<\/span>\r\n    trust_remote_code<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">,<\/span>\r\n    enforce_eager<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span>\r\n<span class=\"token punctuation\">)<\/span>\r\nstop_token_ids <span class=\"token operator\">=<\/span> <span class=\"token punctuation\">[<\/span><span class=\"token number\">151329<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">151336<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token number\">151338<\/span><span class=\"token punctuation\">]<\/span>\r\nsampling_params <span class=\"token operator\">=<\/span> SamplingParams<span class=\"token punctuation\">(<\/span>temperature<span class=\"token operator\">=<\/span><span class=\"token number\">0.95<\/span><span class=\"token punctuation\">,<\/span> max_tokens<span class=\"token operator\">=<\/span><span class=\"token number\">1024<\/span><span class=\"token punctuation\">,<\/span> stop_token_ids<span class=\"token operator\">=<\/span>stop_token_ids<span class=\"token punctuation\">)<\/span>\r\n\r\ninputs <span class=\"token operator\">=<\/span> tokenizer<span class=\"token punctuation\">.<\/span>apply_chat_template<span class=\"token punctuation\">(<\/span>prompt<span class=\"token punctuation\">,<\/span> tokenize<span class=\"token operator\">=<\/span><span class=\"token boolean\">False<\/span><span class=\"token punctuation\">,<\/span> add_generation_prompt<span class=\"token operator\">=<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">)<\/span>\r\noutputs <span class=\"token operator\">=<\/span> llm<span class=\"token punctuation\">.<\/span>generate<span class=\"token punctuation\">(<\/span>prompts<span class=\"token operator\">=<\/span>inputs<span class=\"token punctuation\">,<\/span> sampling_params<span class=\"token operator\">=<\/span>sampling_params<span class=\"token punctuation\">)<\/span>\r\n\r\n<span class=\"token keyword\">print<\/span><span class=\"token punctuation\">(<\/span>outputs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>outputs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>text<span class=\"token punctuation\">)<\/span>\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h2 id=\"-\">\u5e38\u89c1\u95ee\u9898<\/h2>\n<p><strong>Q: \u5982\u4f55\u5904\u7406\u6a21\u578b\u7684\u5185\u5b58\u5360\u7528\u95ee\u9898\uff1f<\/strong><\/p>\n<p><strong>A:<\/strong>\u00a0\u53ef\u4ee5\u901a\u8fc7\u51cf\u5c11<code>max_model_len<\/code>\u53c2\u6570\u503c\u6216\u589e\u52a0<code>tp_size<\/code>\u53c2\u6570\u503c\u6765\u964d\u4f4e\u5185\u5b58\u5360\u7528\u3002<\/p>\n<p><strong>Q: \u5982\u4f55\u4f18\u5316\u6a21\u578b\u7684\u751f\u6210\u6587\u672c\u8d28\u91cf\uff1f<\/strong><\/p>\n<p><strong>A:<\/strong>\u00a0\u53ef\u4ee5\u901a\u8fc7\u8c03\u6574<code>temperature<\/code>\u53c2\u6570\u548c<code>max_tokens<\/code>\u53c2\u6570\u6765\u4f18\u5316\u751f\u6210\u6587\u672c\u7684\u591a\u6837\u6027\u548c\u957f\u5ea6\u3002<\/p>\n<p><strong>Q: \u5982\u4f55\u5904\u7406\u6a21\u578b\u7684\u8f93\u51fa\u4e0d\u7b26\u5408\u9884\u671f\u7684\u60c5\u51b5\uff1f<\/strong><\/p>\n<p><strong>A:<\/strong>\u00a0\u68c0\u67e5\u8f93\u5165\u63d0\u793a\u662f\u5426\u51c6\u786e\uff0c\u6216\u8005\u5c1d\u8bd5\u8c03\u6574\u751f\u6210\u53c2\u6570\u5982<code>do_sample<\/code>\u548c<code>top_k<\/code>\u3002<\/p>\n<h2 id=\"-\">\u76f8\u5173\u8d44\u6e90<\/h2>\n<ul>\n<li><strong>ModelScope\u6a21\u578b\u5e93<\/strong>:\u00a0<a target=\"_blank\" rel=\"noopener noreferrer nofollow\" href=\"https:\/\/www.modelscope.cn\/models\/ZhipuAI\/glm-4-9b-chat\">GLM-4-9B Chat<\/a><\/li>\n<li><strong>GitHub\u4ed3\u5e93<\/strong>: \u66f4\u591a\u63a8\u7406\u4ee3\u7801\u548c\u4f9d\u8d56\u4fe1\u606f\uff0c\u8bf7\u8bbf\u95eeGLM-4\u7684<a target=\"_blank\" rel=\"noopener noreferrer nofollow\" href=\"https:\/\/github.com\/THUDM\/GLM-4\">GitHub<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>GLM-4-9B-Chat\u662f\u667a\u8c31AI\u63a8\u51fa\u7684\u6700\u65b0\u4e00\u4ee3\u9884\u8bad\u7ec3\u6a21\u578bGLM-4\u7cfb\u5217\u4e2d\u7684\u5f00\u6e90\u7248\u672c\u3002\u8be5\u6a21\u578b\u5728\u8bed\u4e49\u7406\u89e3\u3001\u6570\u5b66\u63a8\u7406\u3001\u4ee3\u7801\u6267\u884c\u548c\u77e5\u8bc6\u83b7\u53d6\u7b49\u591a\u4e2a\u65b9\u9762\u8868\u73b0\u51fa\u5353\u8d8a\u7684\u6027\u80fd\u3002\u5b83\u4e0d\u4ec5\u80fd\u591f\u8fdb\u884c\u591a\u8f6e\u5bf9\u8bdd\uff0c\u8fd8\u5177\u5907\u7f51\u9875\u6d4f\u89c8\u3001\u4ee3\u7801\u6267\u884c\u3001\u81ea\u5b9a\u4e49\u5de5\u5177\u8c03\u7528\u548c\u957f\u6587\u672c\u63a8\u7406\u7b49\u9ad8\u7ea7\u529f\u80fd\u3002\u6a21\u578b\u652f\u6301\u5305\u62ec\u65e5\u8bed\u3001\u97e9\u8bed\u3001\u5fb7\u8bed\u5728\u5185\u768426\u79cd\u8bed\u8a00\uff0c\u8fdb\u4e00\u6b65\u589e\u5f3a\u4e86\u5176\u591a\u8bed\u8a00\u5904\u7406\u80fd\u529b\u3002 \u6a21\u578b\u8bc4\u6d4b GLM-4-9B Chat\u5728\u591a\u4e2a\u7ecf\u5178\u4efb\u52a1\u4e0a\u8fdb\u884c\u4e86\u8bc4\u6d4b\uff0c\u7ed3\u679c\u663e\u793a\u5176\u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0a\u5747\u6709\u51fa\u8272\u7684\u8868\u73b0\u3002\u4ee5\u4e0b\u662f\u4e00\u4e9b\u5173\u952e\u8bc4\u6d4b\u7ed3\u679c\uff1a AlignBench-v2: 6.61 MT-Bench: 8<\/p>\n","protected":false},"author":1,"featured_media":178,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"collection":[52],"company":[17],"rank":[],"class_list":["post-177","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-llm","collection-glm-4","company-zhipuai"],"_links":{"self":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/177","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=177"}],"version-history":[{"count":1,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/177\/revisions"}],"predecessor-version":[{"id":179,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/posts\/177\/revisions\/179"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/media\/178"}],"wp:attachment":[{"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/media?parent=177"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/categories?post=177"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/tags?post=177"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/collection?post=177"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/company?post=177"},{"taxonomy":"rank","embeddable":true,"href":"https:\/\/www.onetts.com\/ai\/wp-json\/wp\/v2\/rank?post=177"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}