{"id":1267,"date":"2024-04-30T13:32:19","date_gmt":"2024-04-30T13:32:19","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1267"},"modified":"2024-04-30T13:33:26","modified_gmt":"2024-04-30T13:33:26","slug":"tongyiqianwenkaiyuanmoxingbushushiyong","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/tongyiqianwenkaiyuanmoxingbushushiyong.html","title":{"rendered":"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528"},"content":{"rendered":"

\u9996\u5148\u53ef\u4ee5\u53c2\u8003modelScope\u793e\u533a\u7ed9\u51fa\u7684\u4f7f\u7528\u6587\u6863\uff0c\u5df2\u7ecf\u8db3\u591f\u5168\u9762<\/p>\n

modelscope.cn\/models\/qwen\u2026<\/a><\/p>\n

\u4f46\u5728\u6309\u7167\u6587\u6863\u4e2d\u6b65\u9aa4\u90e8\u7f72\u65f6\uff0c\u8fd8\u662f\u6709\u4e9b\u9519\u8bef\u95ee\u9898\u53d1\u751f\uff0c\u53ef\u4ee5\u641c\u7d22\u53c2\u8003\u7684\u89e3\u51b3\u65b9\u5f0f\u4e0d\u591a\uff0c\u6240\u4ee5\u8bb0\u5f55\u4e0b\u6765<\/p>\n

\u4e2a\u4eba\u7535\u8111\u90e8\u7f72<\/h3>\n

\u8fd9\u91cc\u4e0d\u592a\u5efa\u8bae\u4f7f\u7528\u81ea\u5df1\u7684\u7b14\u8bb0\u672c\u90e8\u7f72\u901a\u4e49\u5343\u95ee\u6a21\u578b\uff0c\u56e0\u4e3a\u5b9e\u5728\u662f\u592a\u8017\u8d44\u6e90\uff0c\u6211\u4f7f\u7528\u7684M2\u82af\u7247\u7684MacBook Pro\u5373\u4f7f\u8fd0\u884c\u8d77\u6765\u4e86\uff0c\u4f46\u6a21\u578b\u56de\u7b54\u4e00\u4e2a\u95ee\u9898\u90fd\u9700\u8981\u56db\u4e94\u5206\u949f\u7684\u65f6\u95f4\uff0c\u5185\u5b58\u5168\u90e8\u5360\u6ee1\uff0c\u5176\u4ed6\u5e94\u7528\u7a0b\u5e8f\u4e5f\u90fd\u5f3a\u5236\u9000\u51fa\u4e86\u3002\u6240\u4ee5\u8fd8\u662f\u4f7f\u7528\u793e\u533a\u63d0\u4f9b\u7684\u514d\u8d39\u8d44\u6e90\uff0c\u6216\u8005\u6709\u66f4\u9ad8\u914d\u7f6e\u7684\u670d\u52a1\u5668\u6765\u90e8\u7f72\u6a21\u578b\u3002\u800c\u4e14\u671f\u95f4\u8fd8\u6709\u5404\u79cd\u95ee\u9898\uff0c\u641c\u4e86\u5f88\u591agithub\u4e0a\u7684\u95ee\u7b54\u624d\u89e3\u51b3\uff0c\u8017\u65f6\u8017\u529b\uff0c\u8fd9\u91cc\u5c31\u4e0d\u8bb0\u5f55\u4e86\uff0c\u5f88\u4e0d\u63a8\u8350\u8fd9\u79cd\u65b9\u5f0f\u3002<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u514d\u8d39\u7b97\u529b\u670d\u52a1\u5668<\/h3>\n

\u6253\u5f00modelScope\u793e\u533a\u540e\uff0c\u70b9\u51fb\u767b\u5f55\u6ce8\u518c\u53ef\u4ee5\u770b\u5230\u514d\u8d39\u8d60\u9001\u7b97\u529b\u7684\u6d3b\u52a8<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u6ce8\u518c\u5b8c\u6210\u540e\u5728\u5bf9\u5e94\u6a21\u578b\u91cc\u53ef\u4ee5\u770b\u5230\uff0c\u968f\u65f6\u90fd\u80fd\u542f\u7528\u7684\u670d\u52a1\u5668<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u8fd9\u91ccCPU\u73af\u5883\u7684\u670d\u52a1\u5668\u52c9\u5f3a\u53ef\u4ee5\u8dd1\u8d77\u6765\u6a21\u578b\uff0c\u4f46\u8fd0\u884c\u6548\u679c\u611f\u4eba\uff0c\u800c\u4e14\u914d\u7f6e\u8fc7\u7a0b\u4e2d\u6709\u5404\u79cd\u95ee\u9898\u9700\u8981\u4fee\u6539\uff0c\u800cGPU\u73af\u5883\u542f\u52a8\u6a21\u578b\u53ef\u4ee5\u8bf4\u662f\u975e\u5e38\u6d41\u7545\uff0c\u4f53\u9a8c\u6548\u679c\u4e5f\u5f88\u597d<\/p>\n

CPU\u73af\u5883\u542f\u52a8<\/h3>\n

\u793e\u533a\u63d0\u4f9b\u7684\u670d\u52a1\u5668\u914d\u7f6e\u5df2\u7ecf\u5f88\u9ad8\u4e86\uff0c8\u683832G\uff0c\u4f46\u56e0\u4e3a\u662f\u7eafCPU\u73af\u5883\uff0c\u542f\u52a8\u8fc7\u7a0b\u4e2d\u8fd8\u662f\u6709\u4e9b\u95ee\u9898<\/p>\n

\u5b89\u88c5\u4f9d\u8d56\u5305<\/h4>\n

\u7b2c\u4e00\u884c\u547d\u4ee4\u4e0d\u9700\u8981\u8fd0\u884c\uff0c\u670d\u52a1\u5668\u5df2\u7ecf\u81ea\u5e26\u4e86modelscope\u5305<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u53ea\u9700\u8981\u65b0\u5efa\u4e00\u4e2aTerminal\u7a97\u53e3\u6765\u6267\u884c\u7b2c\u4e8c\u6761\u547d\u4ee4<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u542f\u52a8\u4ee3\u7801<\/h4>\n

\u76f4\u63a5\u8fd0\u884c\u6587\u6863\u63d0\u4f9b\u7684\u4ee3\u7801\u4f1a\u62a5\u9519\uff0c\u8fd9\u91cc\u662f\u56e0\u4e3a\u7eafCPU\u73af\u5883\u5bfc\u81f4\u7684<\/p>\n

\n

\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n

\u9519\u8bef 1<\/h4>\n

RuntimeError: “addmm_impl_cpu_” not implemented for ‘Half’Hide Error Details<\/p>\n

<\/div>
ini<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>RuntimeError: \"addmm_impl_cpu_\" not implemented for 'Half'<\/span>\n---------------------------------------------------------------------------<\/span>\nRuntimeError                              Traceback (most recent call last)<\/span>\nCell In[1]<\/span>, line 8<\/span>\n      5 model<\/span> = AutoModelForCausalLM.from_pretrained(\"qwen\/Qwen-7B-Chat\"<\/span>, revision = 'v1.0.5'<\/span>,device_map=\"auto\"<\/span>, trust_remote_code=True<\/span>,fp16 = True<\/span>).eval()<\/span>\n      6 model.generation_config<\/span> = GenerationConfig.from_pretrained(\"Qwen\/Qwen-7B-Chat\"<\/span>,revision = 'v1.0.5'<\/span>, trust_remote_code=True<\/span>) # \u53ef\u6307\u5b9a\u4e0d\u540c\u7684\u751f\u6210\u957f\u5ea6\u3001top_p\u7b49\u76f8\u5173\u8d85\u53c2<\/span><\/span>\n----> 8 response, history<\/span> = model.chat(tokenizer, \"\u4f60\u597d\"<\/span>, history=None)<\/span>\n      9 print(response)<\/span>\n     10 response, history<\/span> = model.chat(tokenizer, \"\u6d59\u6c5f\u7684\u7701\u4f1a\u5728\u54ea\u91cc\uff1f\"<\/span>, history=history) <\/span>\n<\/span>\nFile ~\/.cache\/huggingface\/modules\/transformers_modules\/Qwen-7B-Chat\/modeling_qwen.py:1010, in QWenLMHeadModel.chat(self, tokenizer, query, history, system, append_history, stream, stop_words_ids, **kwargs)<\/span>\n   1006 stop_words_ids.extend(get_stop_words_ids(<\/span>\n   1007     self.generation_config.chat_format, tokenizer<\/span>\n   1008 ))<\/span>\n   1009 input_ids<\/span> = torch.tensor([context_tokens]).to(self.device)<\/span>\n-> 1010 outputs<\/span> = self.generate(<\/span>\n   1011             input_ids,<\/span>\n   1012             stop_words_ids<\/span> = stop_words_ids,<\/span>\n   1013             return_dict_in_generate<\/span> = False<\/span>,<\/span>\n   1014             **kwargs,<\/span>\n   1015         )<\/span>\n   1017 response<\/span> = decode_tokens(<\/span>\n   1018     outputs[0]<\/span>,<\/span>\n   1019     tokenizer,<\/span>\n   (...)<\/span>\n   1024     errors<\/span>='replace'<\/span><\/span>\n   1025 )<\/span>\n   1027 if append_history:<\/span>\n<\/code><\/pre>\n
\u9519\u8bef 2<\/h4>\n
ValueError: The current device_map<\/code> had weights offloaded to the disk. Please provide an offload_folder<\/code> for them. Alternatively, make sure you have safetensors<\/code> installed if the model you are using offers the weights in this format.Hide Error Details<\/p>\n
<\/div>
ini<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>ValueError: The current `device_map` had weights offloaded to the disk. Please provide an `offload_folder` for them. Alternatively, make sure you have `safetensors` installed if the model you are using offers the weights in this format.<\/span>\n---------------------------------------------------------------------------<\/span>\nValueError                                Traceback (most recent call last)<\/span>\nCell In[2]<\/span>, line 5<\/span>\n      2 from modelscope import GenerationConfig<\/span>\n      4 tokenizer<\/span> = AutoTokenizer.from_pretrained(\"qwen\/Qwen-7B-Chat\"<\/span>, revision = 'v1.0.5'<\/span>,trust_remote_code=True<\/span>)<\/span>\n----> 5 model<\/span> = AutoModelForCausalLM.from_pretrained(\"qwen\/Qwen-7B-Chat\"<\/span>, revision = 'v1.0.5'<\/span>,device_map=\"auto\"<\/span>, trust_remote_code=True<\/span>,fp16 = True<\/span>).eval()<\/span>\n      6 model.generation_config<\/span> = GenerationConfig.from_pretrained(\"Qwen\/Qwen-7B-Chat\"<\/span>,revision = 'v1.0.5'<\/span>, trust_remote_code=True<\/span>) # \u53ef\u6307\u5b9a\u4e0d\u540c\u7684\u751f\u6210\u957f\u5ea6\u3001top_p\u7b49\u76f8\u5173\u8d85\u53c2<\/span><\/span>\n      7 model.float()<\/span>\n<\/span>\nFile \/opt\/conda\/lib\/python3.8\/site-packages\/modelscope\/utils\/hf_util.py:98, in get_wrapped_class.<locals>.ClassWrapper.from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)<\/span>\n     95 else:<\/span>\n     96     model_dir<\/span> = pretrained_model_name_or_path<\/span>\n---> 98 model<\/span> = module_class.from_pretrained(model_dir, *model_args,<\/span>\n     99                                      **kwargs)<\/span>\n    100 model.model_dir<\/span> = model_dir<\/span>\n    101 return model<\/span>\n<\/code><\/pre>\n
\u89e3\u51b3\u65b9\u5f0f<\/h3>\n
\u9996\u5148\u786e\u4fddtorch 2.0.1\u7248\u672c\uff0c\u7136\u540e\u5728\u4ee3\u7801\u4e2d\u6dfb\u52a0\u8fd9\u4e24\u884c\uff0c\u5373\u53ef\u8fd0\u884c<\/p>\n
model.float()<\/p>\n
offload_folder=”offload_folder”,<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from<\/span> modelscope import<\/span> AutoModelForCausalLM, AutoTokenizer<\/span>\nfrom<\/span> modelscope import<\/span> GenerationConfig<\/span>\nimport<\/span> datetime<\/span>\nprint<\/span>(\"\u542f\u52a8\u65f6\u95f4\uff1a\"<\/span> + str<\/span>(datetime.datetime.now()))<\/span>\ntokenizer = AutoTokenizer.from_pretrained(\"qwen\/Qwen-7B-Chat\"<\/span>, revision = 'v1.0.5'<\/span>,trust_remote_code=True<\/span>)<\/span>\nmodel = AutoModelForCausalLM.from_pretrained(\"qwen\/Qwen-7B-Chat\"<\/span>, revision = 'v1.0.5'<\/span>,device_map=\"auto\"<\/span>,offload_folder=\"offload_folder\"<\/span>, trust_remote_code=True<\/span>,fp16 = True<\/span>).eval<\/span>()<\/span>\nmodel.generation_config = GenerationConfig.from_pretrained(\"Qwen\/Qwen-7B-Chat\"<\/span>,revision = 'v1.0.5'<\/span>, trust_remote_code=True<\/span>) # \u53ef\u6307\u5b9a\u4e0d\u540c\u7684\u751f\u6210\u957f\u5ea6\u3001top_p\u7b49\u76f8\u5173\u8d85\u53c2<\/span><\/span>\nmodel.float<\/span>()<\/span>\n<\/span>\nprint<\/span>(\"\u5f00\u59cb\u6267\u884c\uff1a\"<\/span> + str<\/span>(datetime.datetime.now()))<\/span>\nresponse, history = model.chat(tokenizer, \"\u4f60\u597d\"<\/span>, history=None<\/span>)<\/span>\nprint<\/span>(response)<\/span>\nprint<\/span>(\"\u7b2c\u4e00\u4e2a\u95ee\u9898\u5904\u7406\u5b8c\u6bd5\uff1a\"<\/span> + str<\/span>(datetime.datetime.now()))<\/span>\nresponse, history = model.chat(tokenizer, \"\u6d59\u6c5f\u7684\u7701\u4f1a\u5728\u54ea\u91cc\uff1f\"<\/span>, history=history) <\/span>\nprint<\/span>(response)<\/span>\nprint<\/span>(\"\u7b2c\u4e8c\u4e2a\u95ee\u9898\u5904\u7406\u5b8c\u6bd5\uff1a\"<\/span> + str<\/span>(datetime.datetime.now()))<\/span>\nresponse, history = model.chat(tokenizer, \"\u5b83\u6709\u4ec0\u4e48\u597d\u73a9\u7684\u666f\u70b9\"<\/span>, history=history)<\/span>\nprint<\/span>(response)<\/span>\nprint<\/span>(\"\u7b2c\u4e09\u4e2a\u95ee\u9898\u5904\u7406\u5b8c\u6bd5\uff1a\"<\/span> + str<\/span>(datetime.datetime.now()))<\/span>\n<\/code><\/pre>\n
\u8fd0\u884c\u8d77\u6765\u4e4b\u540e\u901f\u5ea6\u5b9e\u5728\u611f\u4eba\uff0c\u6bcf\u56de\u7b54\u4e00\u4e2a\u95ee\u9898\u90fd\u9700\u8981 5 \u5206\u949f\u5de6\u53f3\uff0c\u8fd8\u6709\u4e00\u5b9a\u6982\u7387\u76f4\u63a5\u542f\u52a8\u5931\u8d25<\/p>\n
\n
\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n
\u542f\u52a8\u6a21\u578b\u8fc7\u7a0b\u4e2d\u4f1a\u51fa\u73b0\u8fd9\u79cd\u62a5\u9519\uff0c\u70b9\u51fbOK\u91cd\u65b0\u6267\u884c\u5c31\u597d\u4e86\uff0c\u53ef\u80fd\u662f\u670d\u52a1\u5668\u8d1f\u8f7d\u592a\u9ad8<\/p>\n
\n
\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>\n
\n
\"\u901a\u4e49\u5343\u95ee\u5f00\u6e90\u6a21\u578b\u90e8\u7f72\u4f7f\u7528\"<\/figure>\n<\/p>","protected":false},"excerpt":{"rendered":"
\u9996\u5148\u53ef\u4ee5\u53c2\u8003modelScope\u793e\u533a\u7ed9\u51fa\u7684\u4f7f\u7528\u6587\u6863\uff0c\u5df2\u7ecf\u8db3\u591f\u5168\u9762 https:\/\/modelscope.cn\/models\/qwen\/Qwen-7B-Chat\/quickstart \u4f46\u5728\u6309\u7167\u6587\u6863\u4e2d\u6b65\u9aa4<\/p>\n","protected":false},"author":1,"featured_media":3035,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"response | \u670d\u52a1\u5668 | revision - \u5948\u65af\u79d1\u6280\u793e\u533a","rank_math_description":"\u5948\u65af\u79d1\u6280\u793e\u533a\u4e13\u6ce8\u4e8eresponse\u3001\u670d\u52a1\u5668\u3001revision\u3001modelchattokenizer\u3001tokenizer\uff0c\u63d0\u4f9b\u4e13\u4e1a\u7684\u5206\u4eab\u548c\u8ba8\u8bba\u5e73\u53f0\u3002","rank_math_focus_keyword":"response,\u670d\u52a1\u5668,revision,modelchattokenizer,tokenizer","views":"50","footnotes":""},"categories":[3],"tags":[136,126,127,128,129],"collection":[],"class_list":["post-1267","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-136","tag-gpt","tag-ai","tag-128","tag-129"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1267","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=1267"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1267\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/3035"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1267"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1267"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1267"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1267"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}