{"id":1438,"date":"2024-05-12T11:33:51","date_gmt":"2024-05-12T11:33:51","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1438"},"modified":"2024-05-12T11:34:33","modified_gmt":"2024-05-12T11:34:33","slug":"ruhezaiopenaidemoxingshangzuoweidiao","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/ruhezaiopenaidemoxingshangzuoweidiao.html","title":{"rendered":"\u5982\u4f55\u5728OpenAI\u7684\u6a21\u578b\u4e0a\u505a\u5fae\u8c03"},"content":{"rendered":"

OpenAI \u5df2\u7ecf\u652f\u6301\u5728\u5b83\u7684\u6a21\u578b\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u5305\u62ec gpt-3.5-turbo\uff0cgpt-4\u3002\u5728\u8fd9\u91cc\uff0c\u628a\u5982\u4f55\u5fae\u8c03 OpenAI \u7684 GPT-3.5-turbo-1106\u7684\u8fc7\u7a0b\u8bb0\u5f55\u4e0b\u6765\u3002<\/p>\n

\u4e00\u3001\u51c6\u5907\u6570\u636e\u96c6<\/h3>\n

\u5fae\u8c03\u4efb\u4f55\u4eba\u5de5\u667a\u80fd\u6a21\u578b\u7684\u7b2c\u4e00\u6b65\u662f\u51c6\u5907\u8bad\u7ec3\u6570\u636e\u96c6\u3002\u5bf9\u4e8e\u6211\u4eec\u7684\u793a\u4f8b\uff0c\u6211\u4eec\u5c06\u4f7f\u7528\u540d\u4e3a \u7684 CSV \u6587\u4ef6\u3002<\/p>\n

\u6b64\u4f8b\u5b50\u4e2d\uff0ctranslate.csv<\/code> \u5176\u4e2d\u5305\u542b\u4e0e\u4e00\u4e9b\u6e38\u620f\u9886\u57df\u7684\u4e2d\u6587\u7ffb\u8bd1\u5185\u5bb9\u3002 origin<\/code>\u4e3a\u539f\u6587\uff0ctarget<\/code>\u4e3a\u8bd1\u6587<\/p>\n

\n

\"1.png\"<\/figure>\n<\/p>\n

\u7136\u540e\uff0c\u6211\u4eec\u5c06\u6b64 CSV \u6587\u4ef6\u8f6c\u6362\u4e3a\u66f4\u9002\u5408\u8bad\u7ec3 AI \u6a21\u578b\u7684 JSONL\uff08JSON Lines\uff09<\/code>\u683c\u5f0f\u3002\u4e0b\u9762\u7684 Python \u811a\u672c\u8bfb\u53d6 CSV \u6587\u4ef6\u5e76\u5c06\u5176\u8f6c\u6362\u4e3a JSONL \u683c\u5f0f\uff1a<\/p>\n

<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\nimport<\/span> json  <\/span>\n<\/span>\nimport<\/span> pandas as<\/span> pd  <\/span>\n<\/span>\nDEFAULT_SYSTEM_PROMPT = '\u628a\u5728>>>\u548c<<<\u4e2d\u7684\u5185\u5bb9\u7ffb\u8bd1\u6210[[\u548c]]\u4e2d\u7684\u8bed\u8a00 '<\/span>  <\/span>\n<\/span>\ndef<\/span> get_example<\/span>(language, origin, target<\/span>):  <\/span>\n<\/span>\n    return<\/span> {  <\/span>\n<\/span>\n        \"messages\"<\/span>: [  <\/span>\n<\/span>\n            {\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: DEFAULT_SYSTEM_PROMPT},  <\/span>\n<\/span>\n            {\"role\"<\/span>: \"user\"<\/span>, \"content\"<\/span>: f'[[{language}<\/span>]], >>>{origin}<\/span><<<'<\/span>},  <\/span>\n<\/span>\n            {\"role\"<\/span>: \"assistant\"<\/span>, \"content\"<\/span>: target},  <\/span>\n<\/span>\n        ]  <\/span>\n<\/span>\n    }  <\/span>\n<\/span>\nif<\/span> __name__ == \"__main__\"<\/span>:  <\/span>\n<\/span>\n    df = pd.read_csv(\"translate.csv\"<\/span>)  <\/span>\n<\/span>\n    with<\/span> open<\/span>(\"train.jsonl\"<\/span>, \"w\"<\/span>, encoding=\"utf8\"<\/span>) as<\/span> f:  <\/span>\n<\/span>\n        for<\/span> i, row in<\/span> list<\/span>(df.iterrows()):  <\/span>\n<\/span>\n            origin = row[\"origin\"<\/span>]  <\/span>\n<\/span>\n            target = row[\"target\"<\/span>]  <\/span>\n<\/span>\n            print<\/span>(origin)<\/span>\n<\/span>\n            example = get_example('en'<\/span>, origin, target)  <\/span>\n<\/span>\n            example_str = json.dumps(example,ensure_ascii=False<\/span>)  <\/span>\n<\/span>\n            f.write(example_str + \"n\"<\/span>)<\/span>\n<\/span>\n  <\/span>\n<\/span>\n<\/span>\n<\/code><\/pre>\n
\u751f\u6210\u7684jsonl\u4e2d\u7684\u5185\u5bb9\u7c7b\u4f3c<\/p>\n
\n
\"2.png\"<\/figure>\n<\/p>\n
\u5177\u4f53\u6bcf\u4e00\u4e2a\u884c\u7c7b\u4f3c<\/p>\n
\n
\"3<\/figure>\n<\/p>\n
\u4e8c\u3001\u8fd0\u884c\u5fae\u8c03<\/h3>\n
\u4e00\u65e6\u6211\u4eec\u51c6\u5907\u597d\u8bad\u7ec3\u6570\u636e\u96c6\uff0c\u6211\u4eec\u5c31\u53ef\u4ee5\u4f7f\u7528 OpenAI \u7684 API \u7ee7\u7eed\u5fae\u8c03\u6211\u4eec\u7684\u6a21\u578b\uff1a<\/p>\n
1\u3001\u5b89\u88c5openai \u5305<\/h4>\n
\u6ce8\u610f\uff1a\u672c\u6587\u7684\u4f8b\u5b50\u9700\u8981 openai \u7248\u672c > 1.1.0 \u7684\uff0c\u5982\u679c\u592a\u8001\u65e7\u4e86\uff0c\u8bf7\u66f4\u65b0<\/strong><\/p>\n
<\/div>
<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\npip install openai<\/span>\n<\/span>\n<\/code><\/pre>\n
2\u3001\u6267\u884c\u5fae\u8c03\u547d\u4ee4<\/h4>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n  <\/span>\n<\/span>\n<\/span>\nimport<\/span> json<\/span>\n<\/span>\nfrom<\/span> time import<\/span> sleep<\/span>\n<\/span>\nfrom<\/span> openai import<\/span> OpenAI<\/span>\n<\/span>\n<\/span>\nimport<\/span> os<\/span>\n<\/span>\nos.environ['OPENAI_API_KEY'<\/span>]=\"sk-7Vl54m90xxxxxxxxxxxxxxxxxxxxxxx\"<\/span><\/span>\n<\/span>\nclient = OpenAI()<\/span>\n<\/span>\n#client = OpenAI(api_key=\"sk-7Vl54m90xxxxxxxxxxxxx\")    #\u6216\u8005\u901a\u8fc7\u53c2\u6570\u4f20\u5165\u7c7b\u4f3c<\/span><\/span>\n<\/span>\n <\/span>\nmodel_name = 'gpt-3.5-turbo-1106'<\/span><\/span>\n<\/span>\n#\u76d1\u63a7\u4efb\u52a1\u5b8c\u6210\u72b6\u6001<\/span><\/span>\n<\/span>\ndef<\/span> wait_untill_done<\/span>(job_id<\/span>):<\/span>\n<\/span>\n    events = {}<\/span>\n<\/span>\n    while<\/span> True<\/span>:<\/span>\n<\/span>\n        response = client.fine_tuning.jobs.list_events(fine_tuning_job_id=job_id, limit=10<\/span>)<\/span>\n<\/span>\n        print<\/span>('fine tuning, waiting for ...'<\/span>)<\/span>\n<\/span>\n        # collect all events<\/span><\/span>\n<\/span>\n        for<\/span> event in<\/span> response.data:<\/span>\n<\/span>\n            if<\/span> \"data\"<\/span> in<\/span> event and<\/span> event.data:<\/span>\n<\/span>\n              print<\/span>(event.data)<\/span>\n<\/span>\n              events[event.data[\"step\"<\/span>]] = event.data[\"train_loss\"<\/span>]<\/span>\n<\/span>\n        messages = [it.message for<\/span> it in<\/span> response.data]<\/span>\n<\/span>\n        for<\/span> m in<\/span> messages:<\/span>\n<\/span>\n            if<\/span> m.startswith(\"New fine-tuned model created: \"<\/span>):<\/span>\n<\/span>\n                return<\/span> m.split(\"created: \"<\/span>)[1<\/span>], events<\/span>\n<\/span>\n        sleep(10<\/span>)<\/span>\n<\/span>\nif<\/span> __name__ == \"__main__\"<\/span>:<\/span>\n<\/span>\n    response = client.files.create(file=open<\/span>(\"train.jsonl\"<\/span>, \"rb\"<\/span>), purpose=\"fine-tune\"<\/span>)<\/span>\n<\/span>\n    uploaded_id = response.id<\/span><\/span>\n<\/span>\n    print<\/span>('uploaded_id='<\/span>, uploaded_id )<\/span>\n<\/span>\n    print<\/span>(\"Dataset is uploaded\"<\/span>)<\/span>\n<\/span>\n    print<\/span>(\"Sleep 10 seconds...\"<\/span>)<\/span>\n<\/span>\n    sleep(10<\/span>)  # wait until dataset would be prepared<\/span><\/span>\n<\/span>\n   <\/span>\n<\/span>\n    response = client.fine_tuning.jobs.create(training_file=uploaded_id,model=model_name)<\/span>\n<\/span>\n    ft_job_id = response.id<\/span><\/span>\n<\/span>\n    print<\/span>(\"Fine-tune job is started, job_id = \"<\/span>,ft_job_id)<\/span>\n<\/span>\n    new_model_name, events = wait_untill_done(ft_job_id)<\/span>\n<\/span>\n    with<\/span> open<\/span>(\"new_model_name.txt\"<\/span>, \"w\"<\/span>) as<\/span> fp:<\/span>\n<\/span>\n        fp.write(new_model_name)<\/span>\n<\/span>\n    print<\/span>(\"Fine-tune job is success, new model name = \"<\/span>,new_model_name)<\/span>\n<\/span>\n<\/code><\/pre>\n
\u6267\u884c\u6210\u529f\u540e\uff0c\u53ef\u4ee5\u770b\u5230\u7c7b\u4f3c\u8fd9\u6837\u7684\u5185\u5bb9<\/p>\n
<\/div>
erlang<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\nuploaded_id= file-3<\/span>KzFOCxKqfZTZe89m1I40wgA<\/span>\n<\/span>\nDataset is uploaded<\/span>\n<\/span>\nSleep 30<\/span> seconds...<\/span>\n<\/span>\nFine-tune job is started, job_id = ftjob-PiqWqQ6BDPbB9hCELN2B6MbL<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\nfine tuning, waiting for ...<\/span>\n<\/span>\n......<\/span>\n<\/span>\nFine-tune job is success, new model name = ft:gpt-3.5<\/span>-turbo-1106<\/span>:personal::8<\/span>LqhuNgA<\/span>\n<\/span>\n<\/code><\/pre>\n
\u5728\u6267\u884c\u7684\u8fc7\u7a0b\u4e2d\uff0c\u4e5f\u53ef\u4ee5\u4e0a platform\u4e0aplatform.openai.com\/finetune<\/a>\uff0c\u53ef\u4ee5\u770b\u5230\u7c7b\u4f3c\u8fd9\u6837\u7684\u5185\u5bb9\u3002<\/p>\n
\n
\"4.png\"<\/figure>\n<\/p>\n
3\u3001\u8bb0\u4f4f\u5fae\u8c03\u540e\u7684model\u540d<\/h4>\n
\u6210\u529f\u4e4b\u540e\uff0c\u53ef\u4ee5\u770b\u5230\u8f93\u51fa Fine-tune job is success, new model name = ft:gpt-3.5-turbo-1106:personal::8LqhuNgA<\/code>\uff0c\u8bb0\u4f4f\u8fd9\u4e2a\u65b0\u7684model name\u3002<\/p>\n
\u5982\u679c\u4e0d\u5c0f\u5fc3\u5173\u6389\u4e86\uff0c\u4e5f\u53ef\u4ee5\u5728 platform\u4e0a\u67e5\u770b\u3002\u5982<\/p>\n
\n
\"5.png\"<\/figure>\n<\/p>\n
\u4e09\u3001\u4f7f\u7528\u5fae\u8c03\u597d\u7684\u6a21\u578b<\/h3>\n
\u6210\u529f\u5fae\u8c03\u6211\u4eec\u7684\u6a21\u578b\u540e\uff0c\u6211\u4eec\u73b0\u5728\u53ef\u4ee5\u4f7f\u7528\u5b83\u6839\u636e\u7528\u6237\u8f93\u5165\u751f\u6210\u54cd\u5e94\uff1a<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\n<\/span>\nfrom<\/span> openai import<\/span> OpenAI<\/span>\n<\/span>\nimport<\/span> os<\/span>\n<\/span>\nos.environ['OPENAI_API_KEY'<\/span>]=\"sk-7Vl54m90xxxxxxxxxxxxxxxxxxxxxxx\"<\/span><\/span>\n<\/span>\nclient = OpenAI()<\/span>\n<\/span>\n#client = OpenAI(api_key=\"sk-7Vl54m90xxxxxxxxxxxxx\")    #\u6216\u8005\u901a\u8fc7\u53c2\u6570\u4f20\u5165\u7c7b\u4f3c<\/span><\/span>\n<\/span>\n<\/span>\nresponse = client.chat.completions.create(<\/span>\n<\/span>\n  model=\"ft:gpt-3.5-turbo-1106:personal::8LqhuNgA\"<\/span>, ##\u6b64\u5904\u4e3a\u4e0a\u9762\u5fae\u8c03\u597d\u7684\u65b0model<\/span><\/span>\n<\/span>\n  messages=[<\/span>\n<\/span>\n    {\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: \"\u4f60\u662f\u4e00\u4e2a\u8bed\u8a00\u4e13\u5bb6\uff0c\u628a\u5728>>>\u548c<<<\u4e2d\u7684\u5185\u5bb9\u7ffb\u8bd1\u6210[[\u548c]]\u4e2d\u7684\u8bed\u8a00 \"<\/span>},<\/span>\n<\/span>\n    {\"role\"<\/span>: \"user\"<\/span>, \"content\"<\/span>: \"[[en]],>>>\u9009\u62e9\u76ee\u6807\u53cb\u65b9\u82f1\u96c4\u5f00\u59cb\u65bd\u6cd5\uff0c\u4e00\u6bb5\u65f6\u95f4\u540e\u4f20\u9001\u81f3\u76ee\u6807\u4f4d\u7f6e \u65bd\u6cd5\u671f\u95f4\u53f3\u65b9\u82f1\u96c4\u83b7\u5f97\u62a4\u76fe\u503c\uff0c\u5e76\u548c{Hero_149}\u83b7\u5f97\u4f24\u5bb3\u51cf\u514d \u4f20\u9001\u5b8c\u6210\u540e{Hero_149}\u589e\u52a0\u79fb\u52a8\u901f\u5ea6\uff0c\u81ea\u8eab\u5468\u56f4\u4e00\u5b9a\u8303\u56f4\u5185\u53f3\u65b9\u82f1\u96c4\u83b7\u5f97\u7269\u7406\u9632\u5fa1\u548c\u9b54\u6cd5\u9632\u5fa1<<<\"<\/span>}<\/span>\n<\/span>\n  ]<\/span>\n<\/span>\n)<\/span>\n<\/span>\nprint<\/span>(response.choices[0<\/span>].message.content)<\/span>\n<\/span>\n<\/span>\n<\/code><\/pre>\n
\u5f97\u5230\u8fd9\u6837\u7684\u7ed3\u679c<\/p>\n
<\/div>
css<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\nselects a<\/span> teammate and starts channeling, then teleports to<\/span> the target after a<\/span> while. While channeling, he grants a<\/span> shield to<\/span> heroes to<\/span> the right<\/span> and damage reduction to<\/span> himself. After teleporting, his Movement Speed is increased, and heroes in range to<\/span> the right<\/span> gain Physical Defense and Magical Defense.<\/span>\n<\/span>\n<\/code><\/pre>","protected":false},"excerpt":{"rendered":"
OpenAI \u5df2\u7ecf\u652f\u6301\u5728\u5b83\u7684\u6a21\u578b\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u5305\u62ec gpt-3.5-turbo\uff0cgpt-4\u3002\u5728\u8fd9\u91cc\uff0c\u628a\u5982\u4f55\u5fae\u8c03 OpenAI \u7684 GPT-3.5-turbo-1106\u7684\u8fc7\u7a0b\u8bb0\u5f55\u4e0b\u6765\u3002 \u4e00\u3001\u51c6\u5907\u6570\u636e\u96c6 \u5fae<\/p>\n","protected":false},"author":1,"featured_media":8524,"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":[128,129,136,126,127],"collection":[],"class_list":["post-1438","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-128","tag-129","tag-136","tag-gpt","tag-ai"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1438","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=1438"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1438\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/8524"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1438"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1438"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1438"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1438"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}