{"id":1943,"date":"2024-06-03T21:37:45","date_gmt":"2024-06-03T21:37:45","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1943"},"modified":"2024-06-03T21:38:46","modified_gmt":"2024-06-03T21:38:46","slug":"bendehuabushuchatglmhefangcaikenggonglue","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/bendehuabushuchatglmhefangcaikenggonglue.html","title":{"rendered":"\u672c\u5730\u5316\u90e8\u7f72Chatglm\u548c\u9632\u8e29\u5751\u653b\u7565"},"content":{"rendered":"

\u6700\u8fd1\u60f3\u641e\u70b9\u4ec0\u4e48\u4e1c\u897f\u7ec3\u7ec3\u624b\uff0c\u4f20\u7edfcrud\u53c8\u6ca1\u6709\u610f\u4e49\uff0c\u4e8e\u662f\u5c31\u770b\u5230\u4e86\u7ed9\u4ecb\u7ecdAI\u7684\u6587\u7ae0\uff0c\u7136\u540e\u5c31\u6162\u6162\u81ea\u5df1\u6478\u7d22\uff0c\u4ece0\u52301\uff0c\u72ec\u81ea\u90e8\u7f72\u5e94\u7528\u3002<\/p>\n

\u9879\u76ee\u7b80\u4ecb<\/h1>\n

ChatGLM3<\/strong> \u662f\u667a\u8c31AI\u548c\u6e05\u534e\u5927\u5b66 KEG \u5b9e\u9a8c\u5ba4\u8054\u5408\u53d1\u5e03\u7684\u5bf9\u8bdd\u9884\u8bad\u7ec3\u6a21\u578b\u3002ChatGLM3-6B \u662f ChatGLM3 \u7cfb\u5217\u4e2d\u7684\u5f00\u6e90\u6a21\u578b\uff0c\u5728\u4fdd\u7559\u4e86\u524d\u4e24\u4ee3\u6a21\u578b\u5bf9\u8bdd\u6d41\u7545\u3001\u90e8\u7f72\u95e8\u69db\u4f4e\u7b49\u4f17\u591a\u4f18\u79c0\u7279\u6027\u7684\u57fa\u7840\u4e0a\uff0cChatGLM3-6B \u5f15\u5165\u4e86\u5982\u4e0b\u7279\u6027\uff1a<\/p>\n

    \n
  1. \u66f4\u5f3a\u5927\u7684\u57fa\u7840\u6a21\u578b\uff1a<\/strong>  ChatGLM3-6B \u7684\u57fa\u7840\u6a21\u578b ChatGLM3-6B-Base \u91c7\u7528\u4e86\u66f4\u591a\u6837\u7684\u8bad\u7ec3\u6570\u636e\u3001\u66f4\u5145\u5206\u7684\u8bad\u7ec3\u6b65\u6570\u548c\u66f4\u5408\u7406\u7684\u8bad\u7ec3\u7b56\u7565\u3002\u5728\u8bed\u4e49\u3001\u6570\u5b66\u3001\u63a8\u7406\u3001\u4ee3\u7801\u3001\u77e5\u8bc6\u7b49\u4e0d\u540c\u89d2\u5ea6\u7684\u6570\u636e\u96c6\u4e0a\u6d4b\u8bc4\u663e\u793a\uff0c* ChatGLM3-6B-Base \u5177\u6709\u5728 10B \u4ee5\u4e0b\u7684\u57fa\u7840\u6a21\u578b\u4e2d\u6700\u5f3a\u7684\u6027\u80fd<\/em>*\u3002<\/li>\n
  2. \u66f4\u5b8c\u6574\u7684\u529f\u80fd\u652f\u6301\uff1a<\/strong>  ChatGLM3-6B \u91c7\u7528\u4e86\u5168\u65b0\u8bbe\u8ba1\u7684 Prompt \u683c\u5f0f<\/a> \uff0c\u9664\u6b63\u5e38\u7684\u591a\u8f6e\u5bf9\u8bdd\u5916\u3002\u540c\u65f6\u539f\u751f\u652f\u6301\u5de5\u5177\u8c03\u7528<\/a>\uff08Function Call\uff09\u3001\u4ee3\u7801\u6267\u884c\uff08Code Interpreter\uff09\u548c Agent \u4efb\u52a1\u7b49\u590d\u6742\u573a\u666f\u3002<\/li>\n
  3. \u66f4\u5168\u9762\u7684\u5f00\u6e90\u5e8f\u5217\uff1a<\/strong>  \u9664\u4e86\u5bf9\u8bdd\u6a21\u578b ChatGLM3-6B<\/a> \u5916\uff0c\u8fd8\u5f00\u6e90\u4e86\u57fa\u7840\u6a21\u578b ChatGLM3-6B-Base<\/a> \u3001\u957f\u6587\u672c\u5bf9\u8bdd\u6a21\u578b ChatGLM3-6B-32K<\/a> \u548c\u8fdb\u4e00\u6b65\u5f3a\u5316\u4e86\u5bf9\u4e8e\u957f\u6587\u672c\u7406\u89e3\u80fd\u529b\u7684 ChatGLM3-6B-128K<\/a>\u3002\u4ee5\u4e0a\u6240\u6709\u6743\u91cd\u5bf9\u5b66\u672f\u7814\u7a76\u5b8c\u5168\u5f00\u653e<\/strong> \uff0c\u5728\u586b\u5199 \u95ee\u5377<\/a> \u8fdb\u884c\u767b\u8bb0\u540e\u4ea6\u5141\u8bb8\u514d\u8d39\u5546\u4e1a\u4f7f\u7528<\/strong>\u3002<\/li>\n<\/ol>\n

    \u5177\u4f53\u8bf7\u67e5\u770b\u5b98\u65b9GitHub github.com\/THUDM\/ChatG\u2026<\/a><\/p>\n

    \u73af\u5883<\/h1>\n

    python<\/h2>\n

    \u9879\u76ee\u8981\u6c42python\u73af\u5883\u6700\u4f4e3.10\uff0c\u5b89\u88c5\u5b8c\u6210\u540e\u6267\u884c\npython -v<\/code>\u67e5\u770bpython\u7248\u672c<\/p>\n

    conda<\/h2>\n

    \u5b89\u88c5conda<\/p>\n

    \u8fd9\u91cc\u6211\u9009\u7684\u662fminiconda\uff0c\u5177\u4f53\u8bf7\u81ea\u884c\u9009\u62e9docs.conda.io\/en\/latest\/<\/a><\/p>\n

    \u5b89\u88c5\u5b8c\u6210\u4f7f\u7528conda -V <\/code>\u67e5\u770b\u7248\u672c\u53f7\u5373\u53ef\u8868\u793a\u5b89\u88c5\u6210\u529f<\/p>\n

    \n

    \"image.png\"<\/figure>\n<\/p>\n

    \u90e8\u7f72<\/h1>\n

    \u5148\u628a\u9879\u76ee\u62ff\u4e0b\u6765<\/p>\n

    git clone https:\/\/github.com\/THUDM\/ChatGLM3<\/code><\/p>\n

    \u5207\u6362\u5230\u6839\u76ee\u5f55\uff0c\u4f7f\u7528pip\u5b89\u88c5\u4f9d\u8d56<\/p>\n

    <\/div>
    bash<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>cd<\/span> ChatGLM3<\/span>\npip install -r requirements.txt<\/span>\n<\/code><\/pre>\n
    \u5728ChatGLM3basic_demo<\/strong>\u76ee\u5f55\u4e0b\u6267\u884c\u547d\u4ee4<\/p>\n
    <\/div>
    arduino<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>streamlit run web_demo_streamlit.py<\/span>\n<\/code><\/pre>\n
    \u7b2c\u4e00\u6b21\u6267\u884c\u9700\u8981\u4e0b\u8f7d\u5f88\u5927\u7684\u6a21\u578b\u6587\u4ef6\uff0c\u5927\u698211g\u5de6\u53f3<\/p>\n
    \n
    \"dcf3eb17fca9d2eeb56647e31ef336a.png\"<\/figure>\n<\/p>\n
    \u7279\u522b\u6ce8\u610f\uff0c\u6b64\u5904\u9700\u8981\u4f7f\u7528\u9b54\u6cd5\uff08\u975e\u56fd\u5185\u8282\u70b9\uff09\uff0c\u5426\u5219\u4f1a\u62a5\u9519\u3002<\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u542f\u52a8\u6210\u529f\u540e\uff0c\u5982\u679c\u4f60\u7684\u63a7\u5236\u53f0\u51fa\u73b0\u4ee5\u4e0b\u4e00\u53e5\u8bdd\u5c31\u8bf4\u660e\u4f60\u4f7f\u7528\u7684\u662fcpu\u5728\u8fd0\u7b97<\/p>\n
    WARNING:root:Some parameters are on the meta device device because they were offloaded to the cpu and disk.<\/strong><\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u56e0\u4e3aCPU\u8dd1\u592a\u6162\u4e86\uff0c\u6211\u4eec\u5fc5\u987b\u8981\u4f7f\u7528GPU\u8fdb\u884c\u8fd0\u7b97\u3002\n\u5728web_demo_streamlit.py<\/strong>\u6587\u4ef6\u91cc\u9762\u6709\u8fd9\u4e48\u4e00\u6bb5\u4ee3\u7801<\/p>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> get_model<\/span>():<\/span>\n    tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_PATH, trust_remote_code=True<\/span>)<\/span>\n    model = AutoModel.from_pretrained(MODEL_PATH, trust_remote_code=True<\/span>, device_map=\"auto\"<\/span>).eval<\/span>()<\/span>\n    return<\/span> tokenizer, model<\/span>\n<\/code><\/pre>\n
    \u6211\u4eec\u628a\u4ed6\u6539\u9020\u4e00\u4e0b<\/p>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> get_model<\/span>():<\/span>\n    tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_PATH, trust_remote_code=True<\/span>)<\/span>\n    model = AutoModel.from_pretrained(\"THUDM\/chatglm3-6b\"<\/span>, trust_remote_code=True<\/span>,device_map=\"cuda\"<\/span>).quantize(4<\/span>).cuda()<\/span>\n    return<\/span> tokenizer, model<\/span>\n<\/code><\/pre>\n
    \u7136\u540e\u518d\u542f\u52a8\u6211\u4eec\u7684\u9879\u76ee\uff0c\u5fc3\u60f3\u7f8e\u6ecb\u6ecb\u554a\uff0c\u7ec8\u4e8e\u80fd\u8dd1\u4e86\u3002<\/p>\n
    \u4f46\u662f\u4e8b\u60c5\u6ca1\u6709\u60f3\u8c61\u4e2d\u90a3\u4e48\u7b80\u5355\u542f\u52a8\u540e\u8fd8\u662f\u4f1a\u51fa\u73b0WARNING:root:Some parameters are on the meta device device because they were offloaded to the cpu and disk.<\/strong><\/p>\n
    pytorch<\/h3>\n
    \u67e5\u9605\u8d44\u6599\u540e\u53d1\u73b0\uff0c\u662f\u56e0\u4e3a\u6211\u4eec\u7684pytorch\u6ca1\u6709\u5b89\u88c5\u6b63\u786e\u7684\u7248\u672c<\/p>\n
    \u4f7f\u7528python\u4ea4\u4e92\u6a21\u5f0f\u8f93\u5165<\/p>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> torch as<\/span> t    <\/span>\nprint<\/span>(t.cuda.is_available())  <\/span>\nprint<\/span>(t.__version__)<\/span>\n<\/code><\/pre>\n
    \u53ef\u4ee5\u770b\u5230\u8f93\u51fa\u662fFalse\u548cxxx+cpu<\/strong>\u8fd9\u6837\u7684\u6570\u636e<\/p>\n
    \u65e2\u7136\u662f\u9519\u8bef\u7684\u7248\u672c\uff0c\u90a3\u91cd\u65b0\u5b89\u88c5\u5c31\u53ef\u4ee5\u4e86\u5427\u3002\n\u6211\u4eec\u53bb\u5230PyTorch\u5b98\u7f51\u6839\u636e\u6211\u4eec\u7684\u673a\u5668\u8fdb\u884c\u7248\u672c\u5b89\u88c5 pytorch.org\/get-started\u2026<\/a><\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    cuda\u7248\u672c<\/h3>\n
    \u63a7\u5236\u53f0\u8f93\u5165<\/p>\n
    <\/div>
    <\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>nvidia-smi<\/span>\n<\/code><\/pre>\n
    \u67e5\u770bGPU\u4fe1\u606f<\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u53f3\u8fb9\u7684cuda\u662f\u6700\u9ad8\u652f\u6301\u7684\u7248\u672c\uff0c\u5de6\u8fb9\u662fGPU\u9a71\u52a8\u4fe1\u606f<\/p>\n
    \u7136\u540e\u6839\u636e\u4f60\u7684\u9a71\u52a8\u7248\u672c\u5728cuda\u5b98\u7f51\u67e5\u770b\u5bf9\u5e94\u7684cuda\u7248\u672c\ndocs.nvidia.com\/cuda\/cuda-t\u2026<\/a><\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u627e\u5230\u5bf9\u5e94\u7684cuda\u7248\u672c\u4e4b\u540e\u6211\u4eec\u5230<\/p>\n
    developer.nvidia.com\/cuda-toolki\u2026<\/a><\/p>\n
    \u4e0b\u8f7d\u4f60\u7684GPU\u5bf9\u5e94\u7684\u7248\u672c\u53f7\u3002<\/p>\n
    \u5b89\u88c5cuda\u7684\u65f6\u5019\u6ce8\u610f\uff0c\u8fd9\u4e2a\u754c\u9762\u53ea\u662f\u628a\u5b89\u88c5\u7a0b\u5e8f\u89e3\u538b\u5230\u6b64\u76ee\u5f55\uff0c\u771f\u6b63\u7684\u5b89\u88c5\u5728\u540e\u9762\u3002<\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u8fd9\u4e2a\u76ee\u5f55cuda\u5b89\u88c5\u5b8c\u6210\u540e\u4f1a\u81ea\u52a8\u5220\u9664\uff0c\u4f46\u662f\u8fd8\u662f\u4e0d\u5efa\u8bae\u653e\u5728c\u76d8\u3002\n\u7b49\u5f85\u4e00\u6bb5\u65f6\u95f4\u540e\u4f1a\u51fa\u73b0\u4ee5\u4e0b\u7a97\u53e3\u3002<\/p>\n
    \n
    \"69ff129d91bb3855091a5c4ccce45a1.png\"<\/figure>\n<\/p>\n
    \u6b64\u5904\u624d\u662f\u771f\u6b63\u7684cuda\u5b89\u88c5\u8def\u5f84\uff0c\u8fd8\u662f\u4e0d\u8981\u653e\u5728c\u76d8\uff0c\u5b89\u88c5\u8fc7\u7a0b\u5c4f\u5e55\u53ef\u80fd\u4f1a\u95ea\u70c1\uff0c\u8fd9\u5c5e\u4e8e\u6b63\u5e38\u73b0\u8c61\u3002<\/p>\n
    \u5b89\u88c5\u5b8c\u6210\u540e\u63a7\u5236\u53f0\u4f7f\u7528\nnvcc -V<\/code>\u67e5\u770b\u7248\u672c<\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\nok\uff0ccuda\u548cpytorch\u90fd\u5b89\u88c5\u5b8c\u6210\u4e86\uff0c\u6211\u4eec\u6d4b\u8bd5\u4e00\u4e0b\u80fd\u4e0d\u80fd\u4f7f\u7528\u5230cuda\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> torch<\/span>\nprint<\/span>(torch.cuda.is_available())<\/span>\n<\/code><\/pre>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> torch<\/span>\nprint<\/span>(torch.__version__)<\/span>\n<\/code><\/pre>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n
    \u6ce8\u610f\uff0c\u7b14\u8005\u4f7f\u7528conda\u5b89\u88c5pytorch\u540e\u8fd0\u884c\u51fa\u73b0\u4ee5\u4e0b\u60c5\u51b5<\/p>\n
    \n
    \"4c7ef9c29be6b33cb068e5daf7590c6.png\"<\/figure>\n<\/p>\n\u7ed3\u679c\u4e00\u770b\u8fd8\u662fcpu\u7248\u672c\u7684pytorch\uff0c\u4e0d\u77e5\u9053\u662f\u4ec0\u4e48\u60c5\u51b5\uff0c\u6b22\u8fce\u8bc4\u8bba\u533a\u6307\u51fa\u3002\n
    \u518d\u6765\u542f\u52a8\u4e00\u6b21AI\uff0c\u7b49\u5f85\u4e00\u6bb5\u65f6\u95f4\u540e\u4f1a\u81ea\u52a8\u8df3\u8f6c\u5230\u4ee5\u4e0b\u754c\u9762<\/p>\n
    \n
    \"image.png\"<\/figure>\n<\/p>\n \u6b64\u56fe\u4e3a\u5b98\u7f51\u56fe\uff0c~~\u6211\u7684\u5fd8\u8bb0\u622a\u56fe\u4e86~~ \n
    \u6211\u672c\u6765\u4ee5\u4e3a\u6211\u7684\u663e\u5361\u80fd\u8dd1\u8d77\u6765\u7684\uff0c\u4f46\u662f\u8bd5\u4e86\u4e4b\u540e\u53d1\u73b0\uff0c\u4e09\u5206\u949f\u8dd1\u4e00\u4e2a\u5b57\n\n
    \"image.png\"<\/figure>\n\n\u53f3\u8fb9GPU\u4f7f\u7528\u548c\u663e\u5b58\u76f4\u63a5\u7206\u4e86\uff0c\u98ce\u6247\u8f6c\u5f97\u98de\u8d77\uff0c\u4e0b\u6b21\u8fd8\u662f\u53bb\u79df\u4e00\u4e2a\u4e91\u670d\u52a1\u5668\u8dd1\u5427\u3002<\/p>\n
    \u603b\u7ed3<\/h1>\n
    \u867d\u7136\u8dd1\u8d77\u6765\u5f88\u5403\u529b\uff0c\u4f46\u662f\u603b\u7b97\u662f\u8e29\u4e86\u51e0\u4e2a\u5751\u8dd1\u5b8c\u4e86\u4e86\uff0c\u53ea\u662f\u6709\u70b9\u6162\uff08<\/p>\n
    \u670b\u53cb\u4eec\u5b89\u88c5pytorch\u7684\u65f6\u5019\u8981\u6ce8\u610f\u7248\u672c\u4ee5\u53cacuda\u7684\u7248\u672c\u5bf9\u5e94\u5173\u7cfb\uff0c\u5751\u90fd\u662f\u8fd9\u4e24\u4e2a\u6bd4\u8f83\u6df1\u3002<\/p>\n
    \u7b14\u8005\u4e5f\u662f\u7b2c\u4e00\u6b21\u73a9\u8fd9\u4e1c\u897f\uff0c\u6b22\u8fce\u5927\u5bb6\u8bc4\u8bba\u533a\u8ba8\u8bba\u6307\u51fa\u4e0d\u8db3\u3002<\/p>","protected":false},"excerpt":{"rendered":"
    \u6700\u8fd1\u60f3\u641e\u70b9\u4ec0\u4e48\u4e1c\u897f\u7ec3\u7ec3\u624b\uff0c\u4f20\u7edfcrud\u53c8\u6ca1\u6709\u610f\u4e49\uff0c\u4e8e\u662f\u5c31\u770b\u5230\u4e86\u7ed9\u4ecb\u7ecdAI\u7684\u6587\u7ae0\uff0c\u7136\u540e\u5c31\u6162\u6162\u81ea\u5df1\u6478\u7d22\uff0c\u4ece0\u52301\uff0c\u72ec\u81ea\u90e8\u7f72\u5e94\u7528\u3002<\/p>\n","protected":false},"author":1,"featured_media":12294,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"","rank_math_description":"","rank_math_focus_keyword":"","views":"4","footnotes":""},"categories":[3],"tags":[126,127,128,129,136],"collection":[],"class_list":["post-1943","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-gpt","tag-ai","tag-128","tag-129","tag-136"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1943","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/comments?post=1943"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1943\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/12294"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1943"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1943"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1943"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1943"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}