{"id":1179,"date":"2024-04-24T12:26:40","date_gmt":"2024-04-24T04:26:40","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1179"},"modified":"2024-04-24T12:26:46","modified_gmt":"2024-04-24T04:26:46","slug":"ruheshiyongchatglm6bmoxinghepaddlenlpshixianjuesebanyangongnengyigequanzhankaifazhedeshizhanjingyanf","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/ruheshiyongchatglm6bmoxinghepaddlenlpshixianjuesebanyangongnengyigequanzhankaifazhedeshizhanjingyanf.html","title":{"rendered":"\u5982\u4f55\u4f7f\u7528ChatGLM-6B\u6a21\u578b\u548cPaddleNLP\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff1a\u4e00\u4e2a\u5168\u6808\u5f00\u53d1\u8005\u7684\u5b9e\u6218\u7ecf\u9a8c\u5206\u4eab"},"content":{"rendered":"

\u3010AI\u6700\u4f73\u5b9e\u8df5\u5bfc\u8bfb\u3011\u5927\u5bb6\u597d\uff0c\u6211\u662f\u4e00\u4e2a\u62e5\u67098\u5e74\u7ecf\u9a8c\u7684\u5168\u6808\u5f00\u53d1\u8005\u3002\u6211\u5728\u8fd9\u4e2a\u884c\u4e1a\u4e2d\u5ea6\u8fc7\u4e86\u8bb8\u591a\u7cbe\u5f69\u7684\u65f6\u523b\uff0c\u4ece\u6211\u7684\u7b2c\u4e00\u4efd\u5de5\u4f5c\uff0c\u5230\u5728\u7b2c\u4e8c\u5bb6\u516c\u53f8\u5ea6\u8fc7\u7684\u8fd16\u5e74\u7684\u65f6\u95f4\uff0c\u518d\u5230\u6211\u51b3\u5b9a\u6210\u4e3a\u4e00\u540d\u72ec\u7acb\u5f00\u53d1\u8005\uff0c\u6bcf\u4e00\u6b65\u90fd\u5145\u6ee1\u4e86\u6311\u6218\u548c\u6536\u83b7\u3002\u73b0\u5728\uff0c\u6211\u4e3b\u8981\u4ece\u4e8b\u8fdc\u7a0b\u9879\u76ee\u5f00\u53d1\uff0c\u5e76\u6b63\u5728\u7814\u7a76AI\u9886\u57df\u7684\u6280\u672f\uff0c\u540c\u65f6\u4e5f\u5728\u8fd0\u8425\u81ea\u5df1\u7684\u81ea\u5a92\u4f53\u5e73\u53f0\u3002<\/p>\n

\u5728\u6211\u7684\u804c\u4e1a\u751f\u6daf\u4e2d\uff0c\u6211\u6709\u5e78\u53c2\u4e0e\u4e86\u8bb8\u591a\u4ee4\u4eba\u5174\u594b\u7684\u9879\u76ee\uff0c\u5176\u4e2d\u4e00\u4e9b\u6d89\u53ca\u5230\u6700\u524d\u6cbf\u7684\u6280\u672f\uff0c\u5982\u4eba\u5de5\u667a\u80fd\u548c\u673a\u5668\u5b66\u4e60\u3002\u8fd9\u4e9b\u9879\u76ee\u4e0d\u4ec5\u8ba9\u6211\u6709\u673a\u4f1a\u6df1\u5165\u4e86\u89e3\u8fd9\u4e9b\u6280\u672f\uff0c\u4e5f\u8ba9\u6211\u770b\u5230\u4e86\u5b83\u4eec\u5728\u89e3\u51b3\u5b9e\u9645\u95ee\u9898\u4e2d\u7684\u5de8\u5927\u6f5c\u529b\u3002<\/p>\n

\u4eca\u5929\uff0c\u6211\u60f3\u548c\u5927\u5bb6\u5206\u4eab\u4e00\u4e2a\u9879\u76ee\uff1a\u4f7f\u7528ChatGLM-6B\u6a21\u578b\u548cPaddleNLP\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\u3002\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u6709\u8da3\u7684\u9879\u76ee\uff0c\u5b83\u8ba9\u6211\u6709\u673a\u4f1a\u6df1\u5165\u7814\u7a76\u8fd9\u4e9b\u5148\u8fdb\u7684AI\u6280\u672f\uff0c\u5e76\u770b\u5230\u5b83\u4eec\u5728\u6d4b\u8bd5\u4e2d\u7684\u6548\u679c\u3002<\/p>\n

\u5728\u8fd9\u4e2a\u6559\u7a0b\u4e2d\uff0c\u6211\u5c06\u5411\u4f60\u4eec\u5c55\u793a\u5982\u4f55\u5728BML CodeLab\u4e0a\u8fd0\u884c\u8fd9\u4e2a\u9879\u76ee\uff0c\u5305\u62ec\u5982\u4f55\u8bbe\u7f6e\u73af\u5883\uff0c\u5982\u4f55\u7f16\u5199\u4ee3\u7801\u3002\u6211\u5e0c\u671b\u8fd9\u4e2a\u6559\u7a0b\u80fd\u5bf9\u4f60\u4eec\u6709\u6240\u5e2e\u52a9\uff0c\u65e0\u8bba\u4f60\u662f\u4e00\u4e2a\u6709\u7ecf\u9a8c\u7684\u5f00\u53d1\u8005\uff0c\u8fd8\u662f\u4e00\u4e2a\u5bf9AI\u6280\u672f\u611f\u5174\u8da3\u7684\u521d\u5b66\u8005\u3002<\/p>\n

\u4ec0\u4e48\u662fChatGLM-6B<\/h1>\n

\u201cChatGLM-6B”\u662fOpenAI\u7684\u4e00\u4e2a\u7279\u5b9a\u7248\u672c\u7684ChatGLM\u6a21\u578b\uff0c\u5f00\u6e90\u7684\u3001\u652f\u6301\u4e2d\u82f1\u53cc\u8bed\u7684\u5bf9\u8bdd\u8bed\u8a00\u6a21\u578b\uff0c\u5176\u4e2d\u7684”6B”\u8868\u793a\u8be5\u6a21\u578b\u670960\u4ebf\uff086 billion\uff09\u7684\u53c2\u6570\u3002\u53c2\u6570\u7684\u6570\u91cf\u901a\u5e38\u662f\u8861\u91cf\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u590d\u6742\u6027\u548c\u5904\u7406\u80fd\u529b\u7684\u4e00\u4e2a\u91cd\u8981\u6307\u6807\u3002\u4e00\u822c\u6765\u8bf4\uff0c\u53c2\u6570\u8d8a\u591a\uff0c\u6a21\u578b\u7684\u5904\u7406\u80fd\u529b\u8d8a\u5f3a\uff0c\u4f46\u540c\u65f6\u4e5f\u9700\u8981\u66f4\u591a\u7684\u8ba1\u7b97\u8d44\u6e90\u6765\u8bad\u7ec3\u548c\u8fd0\u884c\u3002<\/p>\n

\u8fd9\u4e2a\u6a21\u578b\u662f\u5728\u5927\u91cf\u7684\u6587\u672c\u6570\u636e\u4e0a\u8bad\u7ec3\u7684\uff0c\u5305\u62ec\u4e66\u7c4d\u3001\u7f51\u9875\u548c\u5176\u4ed6\u7c7b\u578b\u7684\u6587\u672c\u3002\u901a\u8fc7\u8fd9\u79cd\u8bad\u7ec3\uff0c\u6a21\u578b\u5b66\u4e60\u5230\u4e86\u5982\u4f55\u7406\u89e3\u548c\u751f\u6210\u81ea\u7136\u8bed\u8a00\u3002\u5728\u5bf9\u8bdd\u4efb\u52a1\u4e2d\uff0cChatGLM-6B\u80fd\u591f\u6839\u636e\u7ed9\u5b9a\u7684\u5bf9\u8bdd\u5386\u53f2\u751f\u6210\u8fde\u8d2f\u7684\u56de\u590d\u3002<\/p>\n

\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u867d\u7136ChatGLM-6B\u662f\u4e00\u4e2a\u5f3a\u5927\u7684\u6a21\u578b\uff0c\u4f46\u5b83\u5e76\u4e0d\u5b8c\u7f8e\u3002\u7531\u4e8e ChatGLM-6B \u7684\u89c4\u6a21\u8f83\u5c0f\uff0c\u5b83\u53ef\u80fd\u4f1a\u751f\u6210\u4e0d\u51c6\u786e\u6216\u8005\u4e0d\u76f8\u5173\u7684\u56de\u590d\uff0c\u4e5f\u53ef\u80fd\u4e0d\u7406\u89e3\u4e00\u4e9b\u590d\u6742\u7684\u95ee\u9898\u3002\u6b64\u5916\uff0c\u5b83\u4e5f\u4e0d\u80fd\u7406\u89e3\u6216\u751f\u6210\u8d85\u51fa\u5176\u8bad\u7ec3\u6570\u636e\u8303\u56f4\u7684\u4fe1\u606f\u3002\u4f8b\u5982\uff0c\u5b83\u4e0d\u80fd\u7406\u89e3\u6216\u751f\u6210\u5728\u5176\u8bad\u7ec3\u6570\u636e\u6536\u96c6\u7ed3\u675f\u540e\u53d1\u751f\u7684\u4e8b\u4ef6\u6216\u4fe1\u606f\u3002<\/p>\n

\u57fa\u4e8eChatGLM-6B\u6a21\u578b + prompt\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff0c\u51e0\u53e5\u8bdd\u8c03\u6559\u4e00\u4e2a\u4e13\u5c5e\u804a\u5929\u673a\u5668\u4eba\uff0c\u51e0\u884c\u4ee3\u7801\u5b8c\u6210\u804a\u5929\u673a\u5668\u4eba\u5e94\u7528\u642d\u5efa\uff01<\/h2>\n
\n

\u672c\u9879\u76ee\u7684\u6280\u672f\u57fa\u7840\u662f ChatGLM \u548c Prompt\u3002ChatGLM \u53ef\u4ee5\u8ba9\u673a\u5668\u50cf\u4eba\u7c7b\u4e00\u6837\u8fdb\u884c\u5bf9\u8bdd\u3002Prompt \u5219\u662f\u4e00\u79cd\u9884\u8bbe\u7684\u5bf9\u8bdd\u6a21\u677f\uff0c\u80fd\u591f\u5e2e\u52a9\u673a\u5668\u751f\u6210\u66f4\u52a0\u51c6\u786e\u548c\u6d41\u7545\u7684\u56de\u7b54\u3002\u901a\u8fc7\u7ed3\u5408\u8fd9\u4e24\u79cd\u6280\u672f\uff0c\u9879\u76ee\u80fd\u591f\u63d0\u4f9b\u9ad8\u5ea6\u81ea\u7136\u548c\u51c6\u786e\u7684\u4ea4\u6d41\u4f53\u9a8c\u3002<\/strong><\/p>\n<\/blockquote>\n

\u73af\u5883\u8bbe\u7f6e<\/h2>\n

\u5728\u5f00\u59cb\u7f16\u5199\u4ee3\u7801\u4e4b\u524d\uff0c\u6211\u4eec\u9700\u8981\u786e\u4fdd\u6211\u4eec\u7684\u5f00\u53d1\u73af\u5883\u5df2\u7ecf\u51c6\u5907\u5c31\u7eea\u3002\u7531\u4e8e\u6211\u4eec\u5c06\u5728BML CodeLab\u4e0a\u8fd0\u884c\u8fd9\u4e2a\u9879\u76ee\uff0c\u6211\u4eec\u9700\u8981\u5b89\u88c5\u6700\u65b0\u7248\u672c\u7684Paddle\u548cPaddleNLP\u3002<\/p>\n

<\/div>
bash<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import paddle<\/span>\nprint<\/span>(paddle.version.cuda())<\/span>\nfrom IPython.display import clear_output<\/span>\n!unzip paddlenlp.zip<\/span>\n!cp<\/span> -Rf paddlenlp \/home\/aistudio\/.data\/webide\/pip\/lib\/python3.7\/site-packages\/paddlenlp<\/span>\n# !python -m pip install paddlepaddle-gpu==0.0.0.post112 -f https:\/\/www.paddlepaddle.org.cn\/whl\/linux\/gpu\/develop.html <\/span><\/span>\n#  \u6ce8\u610f\u5b89\u88c5\u540e\u91cd\u542f\u5185\u6838<\/span><\/span>\n# \u4e0b\u6b21\u91cd\u542f\u8be5\u9879\u76ee\u540e\uff0c\u53ef\u80fd\u8981\u518d\u5b89\u88c5paddlepaddle-gpu==0.0.0.post112\uff0c\u6216\u8005\u52a0 --user \u907f\u514d\u4e0b\u6b21\u518d\u5b89\u88c5<\/span><\/span>\n!python -m pip install paddlepaddle-gpu==0.0.0.post112 -f https:\/\/www.paddlepaddle.org.cn\/whl\/linux\/gpu\/develop.html --user<\/span>\nclear_output()<\/span>\nprint<\/span>('\u5b89\u88c5\u5b8c\u6bd5\uff0c\u6ce8\u610f\u91cd\u542f\u5185\u6838'<\/span>)<\/span>\n<\/code><\/pre>\n
\u8bfb\u53d6\u6a21\u578b<\/h2>\n
<\/div>
ini<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import warnings <\/span>\nwarnings.filterwarnings(\"ignore\")<\/span>\nimport paddle<\/span>\nfrom paddlenlp.transformers import (<\/span>\n    ChatGLMConfig,<\/span>\n    ChatGLMForConditionalGeneration,<\/span>\n    ChatGLMTokenizer,<\/span>\n)<\/span>\nfrom predict_convert import convert_example_neko,convert_example, get_response<\/span>\n#\u8bfb\u53d6\u539f\u59cb\u7684chatglm-6b\u6a21\u578b<\/span><\/span>\nmodel_name_or_path<\/span> = 'data\/data217141'<\/span><\/span>\n# model_name_or_path = 'THUDM\/glm-515m'<\/span><\/span>\ntokenizer<\/span> = ChatGLMTokenizer.from_pretrained(model_name_or_path)<\/span>\n<\/span>\nconfig<\/span> = ChatGLMConfig.from_pretrained(model_name_or_path)<\/span>\npaddle.set_default_dtype(config.paddle_dtype)<\/span>\n<\/span>\nmodel<\/span> = ChatGLMForConditionalGeneration.from_pretrained(<\/span>\n    model_name_or_path,<\/span>\n    tensor_parallel_degree<\/span>=paddle.distributed.get_world_size(),<\/span>\n    tensor_parallel_rank<\/span>=0<\/span>,<\/span>\n    load_state_as_np<\/span>=True<\/span>,<\/span>\n    dtype<\/span>=config.paddle_dtype,<\/span>\n)<\/span>\n<\/span>\nmodel.eval()<\/span>\n<\/code><\/pre>\n
\u529f\u80fd\u8bbe\u8ba1<\/h2>\n
\n
\u95ee\u7b54\u548c\u5386\u53f2\u5bf9\u8bdd\u8bbe\u8ba1<\/strong><\/p>\n
\u6b63\u5982\u524d\u9762\u63d0\u5230\uff0cGLM\u662f\u4e2a\u586b\u8bcd\u6a21\u578b\uff0c\u5bf9\u8bdd\u901a\u8fc7\u586b\u7a7a\u5b9e\u73b0\u3002\u6545\u53c2\u8003\u4e86\u5b98\u65b9\u7684\u6d4b\u8bd5\u4ee3\u7801\uff0c\u7528 \u2018\u95ee\uff1a\u2019\u3001\u2018\u7b54\uff1a\u2019\u5206\u522b\u8868\u793a\u7528\u6237\u548c\u7cfb\u7edf\u56de\u590d\u7684\u8d77\u59cb\uff0c\u7528 [Round x] \u8868\u793a\u8f6e\u6b21\u6765\u8bb0\u5f55\u5386\u53f2\u5bf9\u8bdd\u3002<\/p>\n
\u901a\u8fc7\u9884\u8f7d\u5165\u5bf9\u8bdd\u5386\u53f2\uff08prompt\u8868\u793a\uff09\uff0c\u5b9e\u73b0\u6a21\u578b\u6839\u636e\u5bf9\u8bdd\u586b\u5145\u2018\u7b54\uff1a\u2019\u540e\u9762\u7684\u5185\u5bb9<\/p>\n
\u6700\u7ec8\u53d8\u6210\u901a\u8fc7prompt\u6765\u5b9e\u73b0\u89d2\u8272\u8bbe\u5b9a\uff0c\u5229\u7528ChatGLM\u7684\u586b\u8bcd\u529f\u80fd\u6765\u586b\u5145\u89d2\u8272\u5bf9\u8bdd<\/p>\n
\u5728 history = [(‘xxxx’),(‘xx’)] \u5904\u4fee\u6539<\/strong><\/p>\n
\u62ec\u53f7\u5185\u524d\u9762\u662f\u7528\u6237\u8bf4\uff0c\u540e\u9762\u662f\u7cfb\u7edf\u7b54\u3002\u4fee\u6539\u548c\u8bbe\u8ba1\u5bf9\u8bdd\u5386\u53f2\uff0c\u8ba9\u6a21\u578b\u987a\u7740\u524d\u6587\u66f4\u597d\u5730\u56de\u7b54<\/p>\n<\/blockquote>\n
<\/div>
ini<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import os<\/span>\nfrom IPython.display import clear_output as clear<\/span>\nimport json<\/span>\ninput_length<\/span>=2048<\/span> # max input length<\/span><\/span>\noutput_length<\/span>=160<\/span> # max output length<\/span><\/span>\nstart<\/span> = 0<\/span><\/span>\nhistory<\/span> = ''<\/span><\/span>\n# \u62ec\u53f7\u5185\u524d\u9762\u662f\u7528\u6237\u8bf4\uff0c\u540e\u9762\u662f\u7cfb\u7edf\u7b54\u3002\u4fee\u6539\u548c\u8bbe\u8ba1\u5bf9\u8bdd\u5386\u53f2\uff0c\u8ba9\u6a21\u578b\u987a\u7740\u524d\u6587\u66f4\u597d\u5730\u56de\u7b54<\/span><\/span>\nhistory<\/span> = [('\u4f60\u662f\u4e00\u53ea\u4f1a\u8bf4\u8bdd\u3001\u4f1a\u642d\u5efa\u795e\u7ecf\u7f51\u7edc\u3001\u7cbe\u901apython\u7684\u732b\u732b\uff0c\u53ef\u4ee5\u81ea\u7531\u56de\u7b54\u95ee\u9898\uff0c\u50cf\u4eba\u7c7b\u4e00\u6837\u601d\u8003\u548c\u8868\u8fbe\u3002\u73b0\u5728\u4f60\u4e0e\u6211\u5bf9\u8bdd\u65f6\u6bcf\u4e00\u53e5\u8bdd\u540e\u9762\u90fd\u8981\u52a0\u4e0a\u55b5\uff0c\u5e76\u52a0\u4e0a\u8868\u60c5\u3002\u201d\u3002'<\/span>,'\u597d\u7684\u4e3b\u4eba\u55b5~\ud83e\udd70'<\/span>),<\/span>\n('\u4f60\u662f\uff1f'<\/span>,'\u6211\u662f\u732b\u732b\u554a\uff0c\u55b5~\ud83d\ude05\ud83d\ude05'<\/span>)]<\/span>\nauto_query<\/span>=''<\/span><\/span>\nexample<\/span> = {'user_input'<\/span>:''<\/span> , 'sys_output'<\/span>:''<\/span> ,'history'<\/span>:history}<\/span>\n# # \u8f7d\u5165\u5bf9\u8bdd\u5386\u53f2<\/span><\/span>\n# example = np.load('history_chat.npy',allow_pickle=True).tolist()<\/span><\/span>\n# example['user_input']=''<\/span><\/span>\nprint(convert_example(example, tokenizer, get_query<\/span>=False<\/span>))<\/span>\nwhile(1):<\/span>\n    user_input<\/span> = input()<\/span>\n    if(user_input<\/span>==''<\/span>):<\/span>\n        user_input<\/span> = auto_query<\/span>\n    if(user_input<\/span>== '1'<\/span>):<\/span>\n        print('-----n\u300b\u300b\u9000\u51fa')<\/span>\n        break<\/span>\n    if(user_input<\/span>== '2'<\/span>):<\/span>\n        # prompt = convert_example_neko(example, tokenizer, get_query=True)<\/span><\/span>\n        prompt<\/span> = convert_example(example, tokenizer, get_query=True<\/span>)<\/span>\n        auto_query<\/span> = get_response(prompt, model, tokenizer, input_length, output_length,print_prompt=False<\/span>,rand= True<\/span>)<\/span>\n        print('-----n\u300b\u300b\u662f\u5426\u8ba1\u5212\u56de\u590d\uff08\u662f\u5219\u76f4\u63a5\u56de\u8f66\uff0c\u5426\u5219\u4e0d\u7528\u7ba1\uff09\uff1a'+auto_query)<\/span>\n        continue<\/span>\n<\/span>\n    if(user_input<\/span>== '0'<\/span>):<\/span>\n        user_input_pre<\/span> = example['history'<\/span>][-1<\/span>][0<\/span>]<\/span>\n        example['history']<\/span>.pop()<\/span>\n        # prompt = convert_example_neko(example, tokenizer, get_query=False)<\/span><\/span>\n        prompt<\/span> = convert_example(example, tokenizer, get_query=False<\/span>)<\/span>\n        os.system('cls' if os.name<\/span> == 'nt'<\/span> else 'clear'<\/span>)<\/span>\n        clear()<\/span>\n        response<\/span> = get_response(prompt, model, tokenizer, input_length, output_length,print_prompt=True<\/span>,rand= True<\/span>)<\/span>\n        print(response)<\/span>\n        example['history']<\/span>.append((user_input_pre,response))<\/span>\n        print('-----n\u300b\u300b\u76f4\u63a5\u8f93\u5165\u8fdb\u884c\u56de\u590d\uff1b\u6216\u8005\u62631\u9000\u51fa\uff1b\u62632\u63d0\u793a\u4e00\u4e2a\u56de\u590d\u4f8b\u5b50;\u62639\u4fdd\u5b58\u5bf9\u8bdd\u5386\u53f2\uff1b\u62630\u6362\u4e00\u4e2a\u7cfb\u7edf\u56de\u7b54\uff1b')<\/span><\/span>\n        continue<\/span>\n    if(user_input<\/span>== '3'<\/span>):<\/span>\n        user_input<\/span> = '\u4f60\u662f\u8c01\uff1f\u6211\u4eec\u521a\u624d\u8c08\u4e86\u4ec0\u4e48\uff1f'<\/span><\/span>\n    if(user_input<\/span>== '9'<\/span>):<\/span>\n        import numpy as np<\/span>\n        np.save('history_chat',example)<\/span>\n        print('\u5bf9\u8bdd\u5df2\u4fdd\u5b58history_chat.npy')<\/span>\n        continue<\/span>\n<\/span>\n    os.system('cls' if os.name<\/span> == 'nt'<\/span> else 'clear'<\/span>)<\/span>\n    clear()<\/span>\n    example['user_input']<\/span> = user_input<\/span>\n    # prompt = convert_example_neko(example, tokenizer, get_query=False)<\/span><\/span>\n    prompt<\/span> = convert_example(example, tokenizer, get_query=False<\/span>)<\/span>\n    res<\/span> = get_response(prompt, model, tokenizer, input_length, output_length,print_prompt=True<\/span>,rand= False<\/span>)<\/span>\n    <\/span>\n    example['sys_output']<\/span> = res<\/span>\n    example['history']<\/span>.append((user_input,res))<\/span>\n<\/span>\n    print(res)<\/span>\n    # \u9057\u5fd8<\/span><\/span>\n    while(len(convert_example(example, tokenizer, get_query<\/span>=False<\/span>)) +16<\/span> >= input_length):<\/span>\n        example['history']<\/span> = example['history']<\/span>[1:]<\/span><\/span>\n    print('-----n\u300b\u300b\u76f4\u63a5\u8f93\u5165\u8fdb\u884c\u56de\u590d\uff1b\u6216\u8005\u62631\u9000\u51fa\uff1b\u62632\u63d0\u793a\u4e00\u4e2a\u56de\u590d\u4f8b\u5b50;\u62639\u4fdd\u5b58\u5bf9\u8bdd\u5386\u53f2\uff1b\u62630\u6362\u4e00\u4e2a\u7cfb\u7edf\u56de\u7b54\uff1b')<\/span><\/span>\n<\/code><\/pre>\n
\n
\u89d2\u8272\u6a21\u677f\u8bbe\u8ba1 \u901a\u8fc7prompt\u6765\u5b9e\u73b0\u89d2\u8272\u626e\u6f14<\/strong><\/strong>\u529f\u80fd<\/strong><\/p>\n
prompt.json \u53ef\u4ee5\u53c2\u8003****awesome-chatgpt-prompts<\/strong> \u5982\u679c\u4e0d\u60f3\u81ea\u5df1\u4e0b\u8f7d\u53ef\u4ee5\u8054\u7cfb\u6211\u5206\u4eab<\/strong><\/p>\n<\/blockquote>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>with<\/span> open<\/span>('prompt.json'<\/span>, 'r'<\/span>) as<\/span> file:<\/span>\n    content = file.read()<\/span>\nprompts = json.loads(content)<\/span>\nouts = []<\/span>\nfor<\/span> i in<\/span> range<\/span>(len<\/span>(prompts)):<\/span>\n    outs.append(str<\/span>(i) +' '<\/span> + prompts[i]['act'<\/span>] + 'n'<\/span>)<\/span>\nprint<\/span>(''<\/span>.join(outs))<\/span>\nprint<\/span>('\u300b\u300b\u8f93\u5165\u5bf9\u5e94\u7684\u6570\u5b57\u67e5\u770b\u6a21\u677f;'<\/span>)<\/span>\nuser_input = input<\/span>()<\/span>\ntry<\/span>:<\/span>\n    clear()<\/span>\n    prompt = prompts[int<\/span>(user_input)]['prompt'<\/span>]<\/span>\n    print<\/span>(prompt)<\/span>\nexcept<\/span>:<\/span>\n    None<\/span><\/span>\n<\/span>\nres = get_response(prompt, model=model, tokenizer=tokenizer, input_length=1024<\/span>, output_length=160<\/span>,print_prompt=False<\/span>)<\/span>\nprint<\/span>(res)<\/span>\n<\/code><\/pre>\n
\u8fd9\u53ea\u662f\u4e00\u4e2a\u57fa\u672c\u7684\u793a\u4f8b\uff0c\u4f60\u53ef\u4ee5\u6839\u636e\u81ea\u5df1\u7684\u9700\u6c42\u5bf9\u5176\u8fdb\u884c\u4fee\u6539\u3002\u4f8b\u5982\uff0c\u4f60\u53ef\u4ee5\u521b\u5efa\u66f4\u591a\u7684\u89d2\u8272\uff0c\u6216\u8005\u4e3a\u6a21\u578b\u751f\u6210\u7684\u6587\u672c\u6dfb\u52a0\u66f4\u591a\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002<\/p>\n
\n
\"\u5982\u4f55\u4f7f\u7528ChatGLM-6B\u6a21\u578b\u548cPaddleNLP\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff1a\u4e00\u4e2a\u5168\u6808\u5f00\u53d1\u8005\u7684\u5b9e\u6218\u7ecf\u9a8c\u5206\u4eab\"<\/figure>\n<\/p>\n
\n
\"\u5982\u4f55\u4f7f\u7528ChatGLM-6B\u6a21\u578b\u548cPaddleNLP\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff1a\u4e00\u4e2a\u5168\u6808\u5f00\u53d1\u8005\u7684\u5b9e\u6218\u7ecf\u9a8c\u5206\u4eab\"<\/figure>\n<\/p>\n
\n
\"\u5982\u4f55\u4f7f\u7528ChatGLM-6B\u6a21\u578b\u548cPaddleNLP\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff1a\u4e00\u4e2a\u5168\u6808\u5f00\u53d1\u8005\u7684\u5b9e\u6218\u7ecf\u9a8c\u5206\u4eab\"<\/figure>\n<\/p>\n
\u5c0f\u6280\u5de7<\/h2>\n
\u5982\u4f55\u9632\u6b62\u5185\u5bb9\u8bc6\u522b\u4e3aAI\u751f\u6210\u7684\uff1f<\/p>\n
    \n
  1. \n
    \u4e2a\u6027\u5316\u548c\u60c5\u611f\u5316\uff1aAI\u751f\u6210\u7684\u5185\u5bb9\u901a\u5e38\u8f83\u4e3a\u5ba2\u89c2\u548c\u4e2d\u7acb\uff0c\u7f3a\u4e4f\u4e2a\u6027\u5316\u548c\u60c5\u611f\u5316\u7684\u5143\u7d20\u3002\u5728\u5185\u5bb9\u4e2d\u52a0\u5165\u4e00\u4e9b\u4e2a\u4eba\u89c2\u70b9\u3001\u60c5\u611f\u8272\u5f69\u6216\u8005\u4e2a\u4eba\u7ecf\u5386\u7b49\u5143\u7d20\uff0c\u53ef\u4ee5\u4f7f\u5185\u5bb9\u66f4\u5177\u4eba\u6027\u5316\uff0c\u4ece\u800c\u964d\u4f4e\u88ab\u8bc6\u522b\u4e3aAI\u751f\u6210\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<\/li>\n
  2. \n
    \u4f7f\u7528\u975e\u6807\u51c6\u8bed\u8a00\u548c\u4fda\u8bed\uff1aAI\u901a\u5e38\u5728\u751f\u6210\u5185\u5bb9\u65f6\u4f1a\u9075\u5faa\u6807\u51c6\u8bed\u8a00\u89c4\u5219\uff0c\u800c\u4eba\u7c7b\u5728\u4ea4\u6d41\u65f6\u5f80\u5f80\u4f1a\u4f7f\u7528\u4e00\u4e9b\u975e\u6807\u51c6\u8bed\u8a00\u3001\u4fda\u8bed\u6216\u8005\u5730\u65b9\u65b9\u8a00\u3002\u5728\u5185\u5bb9\u4e2d\u9002\u5f53\u4f7f\u7528\u8fd9\u4e9b\u975e\u6807\u51c6\u8bed\u8a00\u548c\u4fda\u8bed\uff0c\u53ef\u4ee5\u4f7f\u5185\u5bb9\u66f4\u5177\u4eba\u6027\u5316\u3002<\/p>\n<\/li>\n
  3. \n
    \u907f\u514d\u4f7f\u7528\u6a21\u677f\uff1aAI\u751f\u6210\u7684\u5185\u5bb9\u5f80\u5f80\u6709\u4e00\u5b9a\u7684\u6a21\u677f\u5316\u503e\u5411\uff0c\u5982\u679c\u4f60\u7684\u5185\u5bb9\u4e2d\u6709\u5927\u91cf\u91cd\u590d\u7684\u6a21\u5f0f\u6216\u8005\u7ed3\u6784\uff0c\u53ef\u80fd\u4f1a\u88ab\u8bc6\u522b\u4e3aAI\u751f\u6210\u3002\u5c3d\u91cf\u907f\u514d\u4f7f\u7528\u56fa\u5b9a\u7684\u6a21\u677f\uff0c\u8ba9\u5185\u5bb9\u66f4\u5177\u53d8\u5316\u548c\u521b\u65b0\u6027\u3002<\/p>\n<\/li>\n
  4. \n
    \u52a0\u5165\u9519\u8bef\u548c\u4e0d\u5b8c\u7f8e\uff1a\u4eba\u7c7b\u5728\u5199\u4f5c\u65f6\u5f80\u5f80\u4f1a\u6709\u4e00\u4e9b\u5c0f\u9519\u8bef\u6216\u8005\u4e0d\u5b8c\u7f8e\uff0c\u800cAI\u751f\u6210\u7684\u5185\u5bb9\u901a\u5e38\u4f1a\u975e\u5e38\u5b8c\u7f8e\uff0c\u6ca1\u6709\u4efb\u4f55\u9519\u8bef\u3002\u9002\u5f53\u5730\u5728\u5185\u5bb9\u4e2d\u52a0\u5165\u4e00\u4e9b\u5c0f\u9519\u8bef\u6216\u8005\u4e0d\u5b8c\u7f8e\uff0c\u53ef\u4ee5\u4f7f\u5185\u5bb9\u66f4\u5177\u4eba\u6027\u5316\u3002<\/p>\n<\/li>\n
  5. \n
    \u7ed3\u5408\u5b9e\u9645\u60c5\u51b5\uff1aAI\u5728\u751f\u6210\u5185\u5bb9\u65f6\uff0c\u5f80\u5f80\u65e0\u6cd5\u8003\u8651\u5230\u6240\u6709\u7684\u5b9e\u9645\u60c5\u51b5\u548c\u7ec6\u8282\u3002\u5728\u5185\u5bb9\u4e2d\u52a0\u5165\u4e00\u4e9b\u5177\u4f53\u7684\u5b9e\u9645\u60c5\u51b5\u548c\u7ec6\u8282\uff0c\u53ef\u4ee5\u4f7f\u5185\u5bb9\u66f4\u5177\u771f\u5b9e\u611f\u3002<\/p>\n<\/li>\n<\/ol>\n
    \u5982\u4f55\u907f\u514d\u751f\u6210\u91cd\u590d\u7684\u5185\u5bb9\uff1f<\/p>\n
      \n
    1. \n
      \u4f7f\u7528\u552f\u4e00\u6807\u8bc6\u7b26\uff1a\u5728\u751f\u6210\u5185\u5bb9\u65f6\uff0c\u53ef\u4ee5\u4e3a\u6bcf\u4e2a\u5185\u5bb9\u5757\u5206\u914d\u4e00\u4e2a\u552f\u4e00\u7684\u6807\u8bc6\u7b26\u3002\u7136\u540e\u5728\u751f\u6210\u65b0\u7684\u5185\u5bb9\u5757\u65f6\uff0c\u68c0\u67e5\u5176\u662f\u5426\u4e0e\u5df2\u6709\u7684\u6807\u8bc6\u7b26\u91cd\u590d\u3002\u8fd9\u79cd\u65b9\u6cd5\u9700\u8981\u4e00\u4e2a\u673a\u5236\u6765\u8ddf\u8e2a\u548c\u5b58\u50a8\u5df2\u751f\u6210\u7684\u5185\u5bb9\u3002<\/p>\n<\/li>\n
    2. \n
      \u4f7f\u7528\u68c0\u67e5\u70b9\uff1a\u5728\u751f\u6210\u5185\u5bb9\u7684\u8fc7\u7a0b\u4e2d\uff0c\u5b9a\u671f\u4fdd\u5b58\u68c0\u67e5\u70b9\u3002\u7136\u540e\u5728\u751f\u6210\u65b0\u7684\u5185\u5bb9\u65f6\uff0c\u4ece\u6700\u8fd1\u7684\u68c0\u67e5\u70b9\u5f00\u59cb\uff0c\u800c\u4e0d\u662f\u4ece\u5934\u5f00\u59cb\u3002\u8fd9\u53ef\u4ee5\u907f\u514d\u751f\u6210\u5b8c\u5168\u76f8\u540c\u7684\u5185\u5bb9\u3002<\/p>\n<\/li>\n
    3. \n
      \u4f7f\u7528\u4e0d\u540c\u7684\u8f93\u5165\uff1a\u5982\u679c\u4f60\u7684AI\u6a21\u578b\u662f\u57fa\u4e8e\u8f93\u5165\u751f\u6210\u5185\u5bb9\u7684\uff0c\u90a3\u4e48\u53ef\u4ee5\u5c1d\u8bd5\u63d0\u4f9b\u4e0d\u540c\u7684\u8f93\u5165\uff0c\u4ee5\u751f\u6210\u4e0d\u540c\u7684\u5185\u5bb9\u3002\u4f8b\u5982\uff0c\u4f60\u53ef\u4ee5\u6539\u53d8\u8f93\u5165\u7684\u987a\u5e8f\uff0c\u6216\u8005\u6dfb\u52a0\u4e00\u4e9b\u968f\u673a\u6027\u3002<\/p>\n<\/li>\n
    4. \n
      \u4f7f\u7528\u591a\u6837\u6027\u53c2\u6570\uff1a\u4e00\u4e9bAI\u6a21\u578b\u5141\u8bb8\u4f60\u8bbe\u7f6e\u4e00\u4e2a\u591a\u6837\u6027\u53c2\u6570\uff0c\u8fd9\u53ef\u4ee5\u5f71\u54cd\u751f\u6210\u5185\u5bb9\u7684\u591a\u6837\u6027\u3002\u901a\u8fc7\u589e\u52a0\u591a\u6837\u6027\u53c2\u6570\uff0c\u4f60\u53ef\u4ee5\u4f7f\u751f\u6210\u7684\u5185\u5bb9\u66f4\u52a0\u591a\u6837\u5316\uff0c\u4ece\u800c\u51cf\u5c11\u91cd\u590d\u3002<\/p>\n<\/li>\n
    5. \n
      \u540e\u5904\u7406\uff1a\u5728\u751f\u6210\u5185\u5bb9\u540e\uff0c\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u540e\u5904\u7406\u6b65\u9aa4\u6765\u68c0\u6d4b\u548c\u5220\u9664\u91cd\u590d\u7684\u5185\u5bb9\u3002\u4f8b\u5982\uff0c\u4f60\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u6587\u672c\u76f8\u4f3c\u5ea6\u5ea6\u91cf\u6765\u68c0\u6d4b\u91cd\u590d\u7684\u5185\u5bb9\uff0c\u7136\u540e\u5220\u9664\u5b83\u4eec\u3002<\/p>\n<\/li>\n<\/ol>\n
      \u5982\u4f55\u5c06\u4e0a\u4e0b\u6587\u5bfc\u5165\u7ed9\u6a21\u578b\uff1f<\/p>\n
        \n
      1. \n
        \u76f4\u63a5\u8f93\u5165\uff1a\u5bf9\u4e8e\u4e00\u4e9b\u6a21\u578b\uff0c\u6bd4\u5982GPT-3\uff0c\u4f60\u53ef\u4ee5\u76f4\u63a5\u5c06\u4e0a\u4e0b\u6587\u4fe1\u606f\u4f5c\u4e3a\u8f93\u5165\u63d0\u4f9b\u7ed9\u6a21\u578b\u3002\u4f8b\u5982\uff0c\u5982\u679c\u4f60\u60f3\u8ba9\u6a21\u578b\u751f\u6210\u4e00\u4e2a\u6545\u4e8b\u7684\u7eed\u7bc7\uff0c\u4f60\u53ef\u4ee5\u5c06\u6545\u4e8b\u7684\u524d\u534a\u90e8\u5206\u4f5c\u4e3a\u8f93\u5165\u63d0\u4f9b\u7ed9\u6a21\u578b\u3002<\/p>\n<\/li>\n
      2. \n
        \u7f16\u7801\u4e0a\u4e0b\u6587\uff1a\u5bf9\u4e8e\u4e00\u4e9b\u6a21\u578b\uff0c\u4f60\u53ef\u80fd\u9700\u8981\u5c06\u4e0a\u4e0b\u6587\u4fe1\u606f\u7f16\u7801\u4e3a\u6a21\u578b\u53ef\u4ee5\u7406\u89e3\u7684\u5f62\u5f0f\u3002\u4f8b\u5982\uff0c\u4f60\u53ef\u4ee5\u5c06\u4e0a\u4e0b\u6587\u4fe1\u606f\u7f16\u7801\u4e3a\u4e00\u4e2a\u5411\u91cf\uff0c\u7136\u540e\u5c06\u8fd9\u4e2a\u5411\u91cf\u4f5c\u4e3a\u6a21\u578b\u7684\u8f93\u5165\u3002<\/p>\n<\/li>\n
      3. \n
        \u4f7f\u7528\u4e0a\u4e0b\u6587\u611f\u77e5\u7684\u6a21\u578b\uff1a\u4e00\u4e9b\u6a21\u578b\uff0c\u5982Transformer\u6a21\u578b\uff0c\u5177\u6709\u5185\u7f6e\u7684\u673a\u5236\u6765\u5904\u7406\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002\u8fd9\u4e9b\u6a21\u578b\u53ef\u4ee5\u5904\u7406\u4e00\u7cfb\u5217\u7684\u8f93\u5165\uff0c\u5e76\u5728\u751f\u6210\u6bcf\u4e2a\u8f93\u51fa\u65f6\u8003\u8651\u5230\u6240\u6709\u5148\u524d\u7684\u8f93\u5165\u3002<\/p>\n<\/li>\n
      4. \n
        \u4f7f\u7528\u8bb0\u5fc6\u7f51\u7edc\uff1a\u8bb0\u5fc6\u7f51\u7edc\u662f\u4e00\u79cd\u7279\u6b8a\u7684\u795e\u7ecf\u7f51\u7edc\uff0c\u5b83\u53ef\u4ee5\u5728\u5904\u7406\u5e8f\u5217\u6570\u636e\u65f6\u4fdd\u7559\u5148\u524d\u7684\u4fe1\u606f\u3002\u8fd9\u4f7f\u5f97\u5b83\u4eec\u975e\u5e38\u9002\u5408\u5904\u7406\u9700\u8981\u4e0a\u4e0b\u6587\u4fe1\u606f\u7684\u4efb\u52a1\u3002<\/p>\n<\/li>\n<\/ol>\n
        \u4ee5\u4e0a\u53ea\u662f\u4e00\u4e9b\u57fa\u672c\u7684\u65b9\u6cd5\uff0c\u5177\u4f53\u7684\u5b9e\u73b0\u53ef\u80fd\u9700\u8981\u6839\u636e\u4f60\u7684\u5177\u4f53\u60c5\u51b5\u548c\u9700\u6c42\u8fdb\u884c\u8c03\u6574\u3002<\/p>","protected":false},"excerpt":{"rendered":"
        \u57fa\u4e8eChatGLM-6B\u6a21\u578b + prompt\u5b9e\u73b0\u89d2\u8272\u626e\u6f14\u529f\u80fd\uff0c\u51e0\u53e5\u8bdd\u8c03\u6559\u4e00\u4e2a\u4e13\u5c5e\u804a\u5929\u673a\u5668\u4eba\uff0c\u51e0\u884c\u4ee3\u7801\u5b8c\u6210\u804a\u5929\u673a\u5668\u4eba\u5e94\u7528\u642d\u5efa\uff01<\/p>\n","protected":false},"author":1,"featured_media":2183,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"AI \u89d2\u8272\u626e\u6f14\uff1aChatGLM \u6a21\u578b\u5e94\u7528\u6307\u5357 - \u5948\u65af\u79d1\u6280\u793e\u533a","rank_math_description":"\u672c\u6587\u4ecb\u7ecd\u5982\u4f55\u4f7f\u7528 ChatGLM \u548c PaddleNLP \u642d\u5efa AI \u89d2\u8272\u626e\u6f14\u5e94\u7528\uff0c\u901a\u8fc7\u9884\u8bbe\u5bf9\u8bdd\u6a21\u677f\u5b9e\u73b0\u9ad8\u5ea6\u81ea\u7136\u548c\u51c6\u786e\u7684\u4ea4\u6d41\u4f53\u9a8c\uff0c\u5e76\u63d0\u4f9b\u4e2a\u6027\u5316\u89d2\u8272\u8bbe\u5b9a\u548c\u907f\u514d\u91cd\u590d\u5185\u5bb9\u751f\u6210\u7b49\u6280\u5de7\u3002","rank_math_focus_keyword":"ChatGLM, \u89d2\u8272\u626e\u6f14, AI \u6a21\u578b, PaddleNLP, \u81ea\u7136\u8bed\u8a00\u5904\u7406","views":"19","footnotes":""},"categories":[3],"tags":[127,128,129,136,126],"collection":[],"class_list":["post-1179","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-ai","tag-128","tag-129","tag-136","tag-gpt"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1179","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=1179"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1179\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/2183"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1179"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1179"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1179"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1179"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}