{"id":1764,"date":"2024-05-29T11:38:20","date_gmt":"2024-05-29T11:38:20","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1764"},"modified":"2024-05-29T11:38:34","modified_gmt":"2024-05-29T11:38:34","slug":"jiyuembeddingmoxinggoujianzhinengwendaxitong","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/jiyuembeddingmoxinggoujianzhinengwendaxitong.html","title":{"rendered":"\u57fa\u4e8eEmbedding\u6a21\u578b\u6784\u5efa\u667a\u80fd\u95ee\u7b54\u7cfb\u7edf"},"content":{"rendered":"

\u672c\u90e8\u5206\u901a\u8fc7\u4e00\u4e2a\u7b80\u5355\u7684\u793a\u4f8b\u4ecb\u7ecd\u5982\u4f55\u6293\u53d6\u7f51\u7ad9\uff08\u672c\u4f8b\u4e2d\u4e3a OpenAI \u7f51\u7ad9\uff09\uff0c\u4f7f\u7528Embedding API \u5c06\u6293\u53d6\u7684\u9875\u9762\u8f6c\u5316\u4e3aEmbedding\u5e76\u5b58\u50a8\uff0c\n\u7136\u540e\u521b\u5efa\u4e00\u4e2a\u57fa\u672c\u7684\u641c\u7d22\u529f\u80fd\uff0c\u5141\u8bb8\u7528\u6237\u57fa\u4e8e\u5b58\u50a8\u7684\u4fe1\u606f\u63d0\u95ee\u3002<\/p>\n

\u4f7f\u7528Scrapy\u722c\u53d6\u7f51\u7ad9\u4fe1\u606f<\/h3>\n

\u9996\u5148\uff0c\u5bfc\u5165\u6240\u9700\u7684\u8f6f\u4ef6\u5305\uff0c\u8bbe\u7f6e\u57fa\u672c\u7684 URL\uff0c\u5e76\u5b9a\u4e49 HTMLParser \u7c7b\u3002<\/p>\n

<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> requests<\/span>\nimport<\/span> re<\/span>\nimport<\/span> urllib.request<\/span>\nfrom<\/span> bs4 import<\/span> BeautifulSoup<\/span>\nfrom<\/span> collections import<\/span> deque<\/span>\nfrom<\/span> html.parser import<\/span> HTMLParser<\/span>\nfrom<\/span> urllib.parse import<\/span> urlparse<\/span>\nimport<\/span> os<\/span>\n<\/span>\n# \u7528\u4e8e\u5339\u914dURL\u7684\u6b63\u5219\u8868\u8fbe\u5f0f\u6a21\u5f0f<\/span><\/span>\nHTTP_URL_PATTERN = r'^http[s]*:\/\/.+'<\/span><\/span>\n<\/span>\ndomain = \"openai.com\"<\/span> # <- \u5c06\u8981\u722c\u53d6\u7684\u57df\u540d\u653e\u5728\u8fd9\u91cc<\/span><\/span>\nfull_url = \"https:\/\/openai.com\/\"<\/span> # <- \u5c06\u8981\u722c\u53d6\u7684\u57df\u540d\u4ee5https\u6216http\u5f62\u5f0f\u653e\u5728\u8fd9\u91cc<\/span><\/span>\n<\/span>\n# \u521b\u5efa\u4e00\u4e2a\u7c7b\u6765\u89e3\u6790HTML\u5e76\u83b7\u53d6\u8d85\u94fe\u63a5<\/span><\/span>\nclass<\/span> HyperlinkParser<\/span>(HTMLParser<\/span>):<\/span>\n    def<\/span> __init__<\/span>(self<\/span>):<\/span>\n        super<\/span>().__init__()<\/span>\n        # \u521b\u5efa\u4e00\u4e2a\u5217\u8868\u6765\u5b58\u50a8\u8d85\u94fe\u63a5<\/span><\/span>\n        self.hyperlinks = []<\/span>\n<\/span>\n    # \u91cd\u5199HTMLParser\u7684handle_starttag\u65b9\u6cd5\u6765\u83b7\u53d6\u8d85\u94fe\u63a5<\/span><\/span>\n    def<\/span> handle_starttag<\/span>(self, tag, attrs<\/span>):<\/span>\n        attrs = dict<\/span>(attrs)<\/span>\n<\/span>\n        # \u5982\u679c\u6807\u7b7e\u662f\u951a\u6807\u7b7e\u5e76\u4e14\u6709href\u5c5e\u6027\uff0c\u5219\u5c06href\u5c5e\u6027\u6dfb\u52a0\u5230\u8d85\u94fe\u63a5\u5217\u8868\u4e2d<\/span><\/span>\n        if<\/span> tag == \"a\"<\/span> and<\/span> \"href\"<\/span> in<\/span> attrs:<\/span>\n            self.hyperlinks.append(attrs[\"href\"<\/span>])<\/span>\n<\/code><\/pre>\n
\u5c06 URL \u4f5c\u4e3a\u53c2\u6570\uff0c\u6253\u5f00 URL \u5e76\u8bfb\u53d6 HTML \u5185\u5bb9\u3002\u7136\u540e\uff0c\u8fd4\u56de\u5728\u8be5\u9875\u9762\u4e0a\u627e\u5230\u7684\u6240\u6709\u8d85\u94fe\u63a5\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> requests<\/span>\nimport<\/span> re<\/span>\nimport<\/span> urllib.request<\/span>\nfrom<\/span> bs4 import<\/span> BeautifulSoup<\/span>\nfrom<\/span> collections import<\/span> deque<\/span>\nfrom<\/span> html.parser import<\/span> HTMLParser<\/span>\nfrom<\/span> urllib.parse import<\/span> urlparse<\/span>\nimport<\/span> os<\/span>\n<\/span>\n# \u7528\u4e8e\u5339\u914dURL\u7684\u6b63\u5219\u8868\u8fbe\u5f0f\u6a21\u5f0f<\/span><\/span>\nHTTP_URL_PATTERN = r'^http[s]*:\/\/.+'<\/span><\/span>\n<\/span>\ndomain = \"openai.com\"<\/span> # <- \u5c06\u8981\u722c\u53d6\u7684\u57df\u540d\u653e\u5728\u8fd9\u91cc<\/span><\/span>\nfull_url = \"https:\/\/openai.com\/\"<\/span> # <- \u5c06\u8981\u722c\u53d6\u7684\u57df\u540d\u4ee5https\u6216http\u5f62\u5f0f\u653e\u5728\u8fd9\u91cc<\/span><\/span>\n<\/span>\n# \u521b\u5efa\u4e00\u4e2a\u7c7b\u6765\u89e3\u6790HTML\u5e76\u83b7\u53d6\u8d85\u94fe\u63a5<\/span><\/span>\nclass<\/span> HyperlinkParser<\/span>(HTMLParser<\/span>):<\/span>\n    def<\/span> __init__<\/span>(self<\/span>):<\/span>\n        super<\/span>().__init__()<\/span>\n        # \u521b\u5efa\u4e00\u4e2a\u5217\u8868\u6765\u5b58\u50a8\u8d85\u94fe\u63a5<\/span><\/span>\n        self.hyperlinks = []<\/span>\n<\/span>\n    # \u91cd\u5199HTMLParser\u7684handle_starttag\u65b9\u6cd5\u6765\u83b7\u53d6\u8d85\u94fe\u63a5<\/span><\/span>\n    def<\/span> handle_starttag<\/span>(self, tag, attrs<\/span>):<\/span>\n        attrs = dict<\/span>(attrs)<\/span>\n<\/span>\n        # \u5982\u679c\u6807\u7b7e\u662f\u951a\u6807\u7b7e\u5e76\u4e14\u6709href\u5c5e\u6027\uff0c\u5219\u5c06href\u5c5e\u6027\u6dfb\u52a0\u5230\u8d85\u94fe\u63a5\u5217\u8868\u4e2d<\/span><\/span>\n        if<\/span> tag == \"a\"<\/span> and<\/span> \"href\"<\/span> in<\/span> attrs:<\/span>\n            self.hyperlinks.append(attrs[\"href\"<\/span>])<\/span>\n<\/span>\n<\/code><\/pre>\n
\u6211\u4eec\u7684\u76ee\u6807\u662f\u53ea\u6293\u53d6 OpenAI \u57df\u540d\u4e0b\u7684\u5185\u5bb9\u5e76\u7f16\u5236\u7d22\u5f15\u3002\u4e3a\u6b64\uff0c\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u8c03\u7528 get_hyperlinks<\/code> \u51fd\u6570\u7684\u51fd\u6570\uff0c\u4f46\u8981\u8fc7\u6ee4\u6389\u4e0d\u5c5e\u4e8e\u6307\u5b9a\u57df\u7684\u4efb\u4f55 URL\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u4eceURL\u83b7\u53d6\u540c\u4e00\u57df\u540d\u5185\u7684\u8d85\u94fe\u63a5\u7684\u51fd\u6570<\/span><\/span>\ndef<\/span> get_domain_hyperlinks<\/span>(local_domain, url<\/span>):<\/span>\n    clean_links = []<\/span>\n    for<\/span> link in<\/span> set<\/span>(get_hyperlinks(url)):<\/span>\n        clean_link = None<\/span><\/span>\n<\/span>\n        # \u5982\u679c\u94fe\u63a5\u662f\u4e00\u4e2aURL\uff0c\u68c0\u67e5\u5b83\u662f\u5426\u5728\u540c\u4e00\u57df\u540d\u5185<\/span><\/span>\n        if<\/span> re.search(HTTP_URL_PATTERN, link):<\/span>\n            # \u89e3\u6790URL\u5e76\u68c0\u67e5\u57df\u540d\u662f\u5426\u76f8\u540c<\/span><\/span>\n            url_obj = urlparse(link)<\/span>\n            if<\/span> url_obj.netloc == local_domain:<\/span>\n                clean_link = link<\/span>\n<\/span>\n        # \u5982\u679c\u94fe\u63a5\u4e0d\u662fURL\uff0c\u68c0\u67e5\u5b83\u662f\u5426\u662f\u4e00\u4e2a\u76f8\u5bf9\u94fe\u63a5<\/span><\/span>\n        else<\/span>:<\/span>\n            if<\/span> link.startswith(\"\/\"<\/span>):<\/span>\n                link = link[1<\/span>:]<\/span>\n            elif<\/span> link.startswith(\"#\"<\/span>) or<\/span> link.startswith(\"mailto:\"<\/span>):<\/span>\n                continue<\/span><\/span>\n            clean_link = \"https:\/\/\"<\/span> + local_domain + \"\/\"<\/span> + link<\/span>\n<\/span>\n        if<\/span> clean_link is<\/span> not<\/span> None<\/span>:<\/span>\n            if<\/span> clean_link.endswith(\"\/\"<\/span>):<\/span>\n                clean_link = clean_link[:-1<\/span>]<\/span>\n            clean_links.append(clean_link)<\/span>\n<\/span>\n    # \u8fd4\u56de\u540c\u4e00\u57df\u540d\u5185\u7684\u8d85\u94fe\u63a5\u5217\u8868<\/span><\/span>\n    return<\/span> list<\/span>(set<\/span>(clean_links))<\/span>\n<\/span>\n<\/code><\/pre>\n
crawl<\/code>\u662f\u7f51\u7edc\u641c\u7d22\u4efb\u52a1\u8bbe\u7f6e\u7684\u6700\u540e\u4e00\u6b65\u3002\u5b83\u8ddf\u8e2a\u8bbf\u95ee\u8fc7\u7684 URL\uff0c\u4ee5\u907f\u514d\u91cd\u590d\u8bbf\u95ee\u540c\u4e00\u9875\u9762\uff0c\u56e0\u4e3a\u540c\u4e00\u9875\u9762\u53ef\u80fd\u5728\u7f51\u7ad9\u7684\u591a\u4e2a\u9875\u9762\u4e2d\u90fd\u6709\u94fe\u63a5\u3002\u5b83\u8fd8\u4f1a\u63d0\u53d6\u7f51\u9875\u4e2d\u6ca1\u6709 HTML \u6807\u8bb0\u7684\u539f\u59cb\u6587\u672c\uff0c\u5e76\u5c06\u6587\u672c\u5185\u5bb9\u5199\u5165\u7279\u5b9a\u4e8e\u8be5\u7f51\u9875\u7684\u672c\u5730 .txt \u6587\u4ef6\u4e2d\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> crawl<\/span>(url<\/span>):<\/span>\n    # \u89e3\u6790URL\u5e76\u83b7\u53d6\u57df\u540d<\/span><\/span>\n    local_domain = urlparse(url).netloc<\/span>\n<\/span>\n    # \u521b\u5efa\u4e00\u4e2a\u961f\u5217\u6765\u5b58\u50a8\u5f85\u722c\u53d6\u7684URLs<\/span><\/span>\n    queue = deque([url])<\/span>\n<\/span>\n    # \u521b\u5efa\u4e00\u4e2a\u96c6\u5408\u6765\u5b58\u50a8\u5df2\u7ecf\u770b\u5230\u7684URLs\uff08\u65e0\u91cd\u590d\uff09<\/span><\/span>\n    seen = set<\/span>([url])<\/span>\n<\/span>\n    # \u521b\u5efa\u4e00\u4e2a\u76ee\u5f55\u6765\u5b58\u50a8\u6587\u672c\u6587\u4ef6<\/span><\/span>\n    if<\/span> not<\/span> os.path.exists(\"text\/\"<\/span>):<\/span>\n            os.mkdir(\"text\/\"<\/span>)<\/span>\n<\/span>\n    if<\/span> not<\/span> os.path.exists(\"text\/\"<\/span>+local_domain+\"\/\"<\/span>):<\/span>\n            os.mkdir(\"text\/\"<\/span> + local_domain + \"\/\"<\/span>)<\/span>\n<\/span>\n    # \u521b\u5efa\u4e00\u4e2a\u76ee\u5f55\u6765\u5b58\u50a8csv\u6587\u4ef6<\/span><\/span>\n    if<\/span> not<\/span> os.path.exists(\"processed\"<\/span>):<\/span>\n            os.mkdir(\"processed\"<\/span>)<\/span>\n<\/span>\n    # \u5f53\u961f\u5217\u4e0d\u4e3a\u7a7a\u65f6\uff0c\u7ee7\u7eed\u722c\u53d6<\/span><\/span>\n    while<\/span> queue:<\/span>\n<\/span>\n        # \u4ece\u961f\u5217\u4e2d\u83b7\u53d6\u4e0b\u4e00\u4e2aURL<\/span><\/span>\n        url = queue.pop()<\/span>\n        print<\/span>(url) # \u7528\u4e8e\u8c03\u8bd5\u548c\u67e5\u770b\u8fdb\u5ea6<\/span><\/span>\n<\/span>\n        # \u5c06url\u7684\u6587\u672c\u4fdd\u5b58\u5230<url>.txt\u6587\u4ef6\u4e2d<\/span><\/span>\n        with<\/span> open<\/span>('text\/'<\/span>+local_domain+'\/'<\/span>+url[8<\/span>:].replace(\"\/\"<\/span>, \"_\"<\/span>) + \".txt\"<\/span>, \"w\"<\/span>, encoding=\"UTF-8\"<\/span>) as<\/span> f:<\/span>\n<\/span>\n            # \u4f7f\u7528BeautifulSoup\u4eceURL\u83b7\u53d6\u6587\u672c<\/span><\/span>\n            soup = BeautifulSoup(requests.get(url).text, \"html.parser\"<\/span>)<\/span>\n<\/span>\n            # \u83b7\u53d6\u6587\u672c\u4f46\u53bb\u9664\u6807\u7b7e<\/span><\/span>\n            text = soup.get_text()<\/span>\n<\/span>\n            # \u5982\u679c\u722c\u866b\u5230\u8fbe\u9700\u8981JavaScript\u7684\u9875\u9762\uff0c\u5b83\u5c06\u505c\u6b62\u722c\u53d6<\/span><\/span>\n            if<\/span> (\"You need to enable JavaScript to run this app.\"<\/span> in<\/span> text):<\/span>\n                print<\/span>(\"\u65e0\u6cd5\u89e3\u6790\u9875\u9762 \"<\/span> + url + \" \u56e0\u4e3a\u9700\u8981\u542f\u7528JavaScript\"<\/span>)<\/span>\n<\/span>\n            # \u5426\u5219\uff0c\u5c06\u6587\u672c\u5199\u5165text\u76ee\u5f55\u4e0b\u7684\u6587\u4ef6<\/span><\/span>\n            f.write(text)<\/span>\n<\/span>\n        # \u4eceURL\u83b7\u53d6\u8d85\u94fe\u63a5\u5e76\u5c06\u5b83\u4eec\u6dfb\u52a0\u5230\u961f\u5217\u4e2d<\/span><\/span>\n        for<\/span> link in<\/span> get_domain_hyperlinks(local_domain, url):<\/span>\n            if<\/span> link not<\/span> in<\/span> seen:<\/span>\n                queue.append(link)<\/span>\n                seen.add(link)<\/span>\n<\/span>\ncrawl(full_url)<\/span>\n<\/code><\/pre>\n
\u4e0a\u8ff0\u793a\u4f8b\u7684\u6700\u540e\u4e00\u884c\u8fd0\u884c\u722c\u7f51\u7a0b\u5e8f\uff0c\u8be5\u7a0b\u5e8f\u4f1a\u6d4f\u89c8\u6240\u6709\u53ef\u8bbf\u95ee\u94fe\u63a5\uff0c\u5e76\u5c06\u8fd9\u4e9b\u9875\u9762\u8f6c\u5316\u4e3a\u6587\u672c\u6587\u4ef6\u3002\u6839\u636e\u7f51\u7ad9\u7684\u5927\u5c0f\u548c\u590d\u6742\u7a0b\u5ea6\uff0c\u8fd9\u9700\u8981\u51e0\u5206\u949f\u7684\u65f6\u95f4\u3002<\/p>\n
\u6784\u5efaEmbedding\u7d22\u5f15<\/h3>\n
\u8981\u5c06\u6587\u672c\u8f6c\u6362\u4e3a CSV\uff0c\u9700\u8981\u5faa\u73af\u6d4f\u89c8\u4e4b\u524d\u521b\u5efa\u7684\u6587\u672c\u76ee\u5f55\u4e2d\u7684\u6587\u672c\u6587\u4ef6\u3002\u6253\u5f00\u6bcf\u4e2a\u6587\u4ef6\u540e\uff0c\u5220\u9664\u591a\u4f59\u7684\u884c\u8ddd\uff0c\u5e76\u5c06\u4fee\u6539\u540e\u7684\u6587\u672c\u6dfb\u52a0\u5230\u5217\u8868\u4e2d\u3002\u7136\u540e\uff0c\u5c06\u5220\u9664\u4e86\u65b0\u884c\u7684\u6587\u672c\u6dfb\u52a0\u5230\u4e00\u4e2a\u7a7a\u7684 Pandas \u6570\u636e\u6846\u4e2d\uff0c\u5e76\u5c06\u6570\u636e\u6846\u5199\u5165 CSV \u6587\u4ef6\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> pandas as<\/span> pd<\/span>\n<\/span>\ndef<\/span> remove_newlines<\/span>(serie<\/span>):<\/span>\n    serie = serie.str<\/span>.replace('n'<\/span>, ' '<\/span>)<\/span>\n    serie = serie.str<\/span>.replace('n'<\/span>, ' '<\/span>)<\/span>\n    serie = serie.str<\/span>.replace('  '<\/span>, ' '<\/span>)<\/span>\n    serie = serie.str<\/span>.replace('  '<\/span>, ' '<\/span>)<\/span>\n    return<\/span> serie<\/span>\n<\/span>\n# \u521b\u5efa\u4e00\u4e2a\u5217\u8868\u6765\u5b58\u50a8\u6587\u672c\u6587\u4ef6<\/span><\/span>\ntexts=[]<\/span>\n<\/span>\n# \u83b7\u53d6text\u76ee\u5f55\u4e2d\u7684\u6240\u6709\u6587\u672c\u6587\u4ef6<\/span><\/span>\nfor<\/span> file in<\/span> os.listdir(\"text\/\"<\/span> + domain + \"\/\"<\/span>):<\/span>\n<\/span>\n    # \u6253\u5f00\u6587\u4ef6\u5e76\u8bfb\u53d6\u6587\u672c<\/span><\/span>\n    with<\/span> open<\/span>(\"text\/\"<\/span> + domain + \"\/\"<\/span> + file, \"r\"<\/span>, encoding=\"UTF-8\"<\/span>) as<\/span> f:<\/span>\n        text = f.read()<\/span>\n<\/span>\n        # \u7701\u7565\u524d11\u884c\u548c\u6700\u540e4\u884c\uff0c\u7136\u540e\u5c06-\u3001_\u548c#update\u66ff\u6362\u4e3a\u7a7a\u683c\u3002<\/span><\/span>\n        texts.append((file[11<\/span>:-4<\/span>].replace('-'<\/span>,' '<\/span>).replace('_'<\/span>, ' '<\/span>).replace('#update'<\/span>,''<\/span>), text))<\/span>\n<\/span>\n# \u4ece\u6587\u672c\u5217\u8868\u521b\u5efa\u4e00\u4e2adataframe<\/span><\/span>\ndf = pd.DataFrame(texts, columns = ['fname'<\/span>, 'text'<\/span>])<\/span>\n<\/span>\n# \u5c06text\u5217\u8bbe\u7f6e\u4e3a\u5220\u9664\u4e86\u6362\u884c\u7b26\u7684\u539f\u59cb\u6587\u672c<\/span><\/span>\ndf['text'<\/span>] = df['fname'<\/span>] + \". \"<\/span> + df['text'<\/span>].replace('n'<\/span>, ' '<\/span>)<\/span>\ndf.to_csv('processed\/scraped.csv'<\/span>)<\/span>\ndf.head()<\/span>\n<\/code><\/pre>\n
\u5c06\u539f\u59cb\u6587\u672c\u4fdd\u5b58\u5230 CSV \u6587\u4ef6\u540e\uff0c\u4e0b\u4e00\u6b65\u662fTokenization\u3002\u6b64\u8fc7\u7a0b\u901a\u8fc7\u5206\u89e3\u53e5\u5b50\u548c\u5355\u8bcd\u5c06\u8f93\u5165\u6587\u672c\u62c6\u5206\u4e3atokens\u3002<\/p>\n
\n
\u4e00\u4e2a\u6709\u7528\u7684\u7ecf\u9a8c\u6cd5\u5219\u662f\uff0c\u4e00\u4e2atoken\u901a\u5e38\u5bf9\u5e94\u4e8e\u5e38\u89c1\u82f1\u8bed\u6587\u672c\u7684 4 \u4e2a\u5b57\u7b26\u3002\u76f8\u5f53\u4e8e\u5927\u7ea6 3\/4 \u4e2a\u5355\u8bcd\uff08\u56e0\u6b64 100 \u4e2a\u6807\u8bb0 ~= 75 \u4e2a\u5355\u8bcd\uff09\u3002<\/p>\n<\/blockquote>\n
API \u5bf9Embedding\u7684\u8f93\u5165token\u7684\u6700\u5927\u6570\u91cf\u6709\u9650\u5236\u3002\u4e3a\u4e86\u4fdd\u8bc1\u4e0d\u8d85\u8fc7\u9650\u5236\uff0cCSV \u6587\u4ef6\u4e2d\u7684\u6587\u672c\u9700\u8981\u5206\u89e3\u4e3a\u591a\u884c\u3002\u9996\u5148\u8bb0\u5f55\u6bcf\u884c\u73b0\u6709\u7684\u957f\u5ea6\uff0c\u4ee5\u8bc6\u522b\u54ea\u4e9b\u884c\u9700\u8981\u62c6\u5206\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> tiktoken<\/span>\n<\/span>\n# \u52a0\u8f7dcl100k_base\u5206\u8bcd\u5668\uff0c\u5b83\u8bbe\u8ba1\u7528\u4e8e\u4e0eada-002\u6a21\u578b\u4e00\u8d77\u5de5\u4f5c<\/span><\/span>\ntokenizer = tiktoken.get_encoding(\"cl100k_base\"<\/span>)<\/span>\n<\/span>\ndf = pd.read_csv('processed\/scraped.csv'<\/span>, index_col=0<\/span>)<\/span>\ndf.columns = ['title'<\/span>, 'text'<\/span>]<\/span>\n<\/span>\n# \u5bf9\u6587\u672c\u8fdb\u884c\u5206\u8bcd\uff0c\u5e76\u5c06\u5206\u8bcd\u6570\u4fdd\u5b58\u5230\u65b0\u7684\u5217\u4e2d<\/span><\/span>\ndf['n_tokens'<\/span>] = df.text.apply(lambda<\/span> x: len<\/span>(tokenizer.encode(x)))<\/span>\n<\/span>\n# \u4f7f\u7528\u76f4\u65b9\u56fe\u53ef\u89c6\u5316\u6bcf\u884c\u7684\u5206\u8bcd\u6570\u91cf\u5206\u5e03<\/span><\/span>\ndf.n_tokens.hist()<\/span>\n<\/code><\/pre>\n
\n
\"tokens\u6570\u91cf\u5206\u5e03\"<\/figure>\n\n\u6700\u65b0\u7684Embedding\u6a21\u578b\u53ef\u4ee5\u5904\u7406\u591a\u8fbe 8191 \u4e2a\u8f93\u5165\u6807\u8bb0\u7684\u8f93\u5165\uff0c\u56e0\u6b64\u5927\u90e8\u5206\u884c\u90fd\u4e0d\u9700\u8981\u4efb\u4f55\u5206\u5757\uff0c\u4f46\u5e76\u4e0d\u662f\u6bcf\u4e2a\u5b50\u9875\u9762\u90fd\u9700\u8981\u5206\u5757\uff0c\u56e0\u6b64\u4e0b\u4e00\u4e2a\u4ee3\u7801\u5757\u5c06\u628a\u8f83\u957f\u7684\u884c\u5206\u5272\u6210\u8f83\u5c0f\u7684\u5757\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>max_tokens = 500<\/span><\/span>\n<\/span>\n# \u51fd\u6570\u7528\u4e8e\u5c06\u6587\u672c\u62c6\u5206\u4e3a\u6700\u5927\u4ee4\u724c\u6570\u7684\u591a\u4e2a\u5757<\/span><\/span>\ndef<\/span> split_into_many<\/span>(text, max_tokens = max_tokens<\/span>):<\/span>\n<\/span>\n    # \u5c06\u6587\u672c\u62c6\u5206\u6210\u53e5\u5b50<\/span><\/span>\n    sentences = text.split('. '<\/span>)<\/span>\n<\/span>\n    # \u83b7\u53d6\u6bcf\u4e2a\u53e5\u5b50\u7684\u4ee4\u724c\u6570<\/span><\/span>\n    n_tokens = [len<\/span>(tokenizer.encode(\" \"<\/span> + sentence)) for<\/span> sentence in<\/span> sentences]<\/span>\n<\/span>\n    chunks = []<\/span>\n    tokens_so_far = 0<\/span><\/span>\n    chunk = []<\/span>\n<\/span>\n    # \u904d\u5386\u5143\u7ec4\u4e2d\u8054\u5408\u7684\u53e5\u5b50\u548c\u4ee4\u724c<\/span><\/span>\n    for<\/span> sentence, token in<\/span> zip<\/span>(sentences, n_tokens):<\/span>\n<\/span>\n        # \u5982\u679c\u5230\u76ee\u524d\u4e3a\u6b62\u7684\u4ee4\u724c\u6570\u52a0\u4e0a\u5f53\u524d\u53e5\u5b50\u7684\u4ee4\u724c\u6570\u5927\u4e8e\u6700\u5927\u4ee4\u724c\u6570\uff0c\u5219\u5c06\u5757\u6dfb\u52a0\u5230\u5757\u5217\u8868\u4e2d\uff0c\u5e76\u91cd\u7f6e<\/span><\/span>\n        # \u5757\u548c\u5230\u76ee\u524d\u4e3a\u6b62\u7684\u4ee4\u724c<\/span><\/span>\n        if<\/span> tokens_so_far + token > max_tokens:<\/span>\n            chunks.append(\". \"<\/span>.join(chunk) + \".\"<\/span>)<\/span>\n            chunk = []<\/span>\n            tokens_so_far = 0<\/span><\/span>\n<\/span>\n        # \u5982\u679c\u5f53\u524d\u53e5\u5b50\u7684\u4ee4\u724c\u6570\u5927\u4e8e\u6700\u5927\u4ee4\u724c\u6570\uff0c\u8df3\u5230\u4e0b\u4e00\u53e5<\/span><\/span>\n        if<\/span> token > max_tokens:<\/span>\n            continue<\/span><\/span>\n<\/span>\n        # \u5426\u5219\uff0c\u5c06\u53e5\u5b50\u6dfb\u52a0\u5230\u5757\u4e2d\uff0c\u5e76\u5c06\u4ee4\u724c\u6570\u52a0\u5230\u603b\u6570\u4e0a<\/span><\/span>\n        chunk.append(sentence)<\/span>\n        tokens_so_far += token + 1<\/span><\/span>\n<\/span>\n    return<\/span> chunks<\/span>\n<\/span>\n<\/span>\nshortened = []<\/span>\n<\/span>\n# \u904d\u5386\u6570\u636e\u5e27<\/span><\/span>\nfor<\/span> row in<\/span> df.iterrows():<\/span>\n<\/span>\n    # \u5982\u679c\u6587\u672c\u4e3aNone\uff0c\u8df3\u5230\u4e0b\u4e00\u884c<\/span><\/span>\n    if<\/span> row[1<\/span>]['text'<\/span>] is<\/span> None<\/span>:<\/span>\n        continue<\/span><\/span>\n<\/span>\n    # \u5982\u679c\u4ee4\u724c\u6570\u5927\u4e8e\u6700\u5927\u4ee4\u724c\u6570\uff0c\u5c06\u6587\u672c\u62c6\u5206\u6210\u591a\u4e2a\u5757<\/span><\/span>\n    if<\/span> row[1<\/span>]['n_tokens'<\/span>] > max_tokens:<\/span>\n        shortened += split_into_many(row[1<\/span>]['text'<\/span>])<\/span>\n<\/span>\n    # \u5426\u5219\uff0c\u5c06\u6587\u672c\u6dfb\u52a0\u5230\u7f29\u77ed\u6587\u672c\u5217\u8868\u4e2d<\/span><\/span>\n    else<\/span>:<\/span>\n        shortened.append(row[1<\/span>]['text'<\/span>])<\/span>\n<\/span>\n<\/code><\/pre>\n
\u518d\u6b21\u53ef\u89c6\u5316\u66f4\u65b0\u540e\u7684\u76f4\u65b9\u56fe\u6709\u52a9\u4e8e\u786e\u8ba4\u884c\u662f\u5426\u5df2\u6210\u529f\u62c6\u5206\u4e3a\u7f29\u77ed\u7684\u90e8\u5206\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>df = pd.DataFrame(shortened, columns = ['text'<\/span>])<\/span>\ndf['n_tokens'<\/span>] = df.text.apply(lambda<\/span> x: len<\/span>(tokenizer.encode(x)))<\/span>\ndf.n_tokens.hist()<\/span>\n<\/code><\/pre>\n
\n
\"\u62c6\u5206\u540e\u7684tokens\u6570\u91cf\u5206\u5e03\"<\/figure>\n\n\u73b0\u5728\uff0c\u5185\u5bb9\u88ab\u5206\u89e3\u6210\u66f4\u5c0f\u7684\u5757\uff0c\u53ef\u4ee5\u5411 OpenAI API \u53d1\u9001\u4e00\u4e2a\u7b80\u5355\u7684\u8bf7\u6c42\uff0c\u6307\u5b9a\u4f7f\u7528\u65b0\u7684text-embedding-ada-002  \u6a21\u578b\u6765\u521b\u5efaembedding:<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from<\/span> openai import<\/span> OpenAI<\/span>\n<\/span>\nclient = OpenAI(<\/span>\n    api_key=os.environ.get(\"OPENAI_API_KEY\"<\/span>),<\/span>\n)<\/span>\n<\/span>\ndf['embeddings'<\/span>] = df.text.apply(lambda<\/span> x: client.embeddings.create(input<\/span>=x, engine='text-embedding-ada-002'<\/span>)['data'<\/span>][0<\/span>]['embedding'<\/span>])<\/span>\n<\/span>\ndf.to_csv('processed\/embeddings.csv'<\/span>)<\/span>\ndf.head()<\/span>\n<\/code><\/pre>\n
\u8fd9\u5e94\u8be5\u9700\u8981\u5927\u7ea6 3-5 \u5206\u949f\uff0c\u4f46\u4e4b\u540e\u60a8\u7684\u5d4c\u5165\u5c31\u53ef\u4ee5\u4f7f\u7528\u4e86\uff01<\/p>\n
\u6839\u636eEmbedding\u6784\u5efa\u667a\u80fd\u95ee\u7b54\u7cfb\u7edf<\/h2>\n
\u7b2c\u4e00\u6b65\u662f\u5c06\u5d4c\u5165\u6570\u636e\u8f6c\u5316\u4e3a NumPy \u6570\u7ec4\uff0c\u8fd9\u5c06\u4e3a\u5982\u4f55\u4f7f\u7528\u5d4c\u5165\u6570\u636e\u63d0\u4f9b\u66f4\u591a\u7075\u6d3b\u6027\uff0c\u56e0\u4e3a\u6709\u8bb8\u591a\u51fd\u6570\u53ef\u4ee5\u5bf9 NumPy \u6570\u7ec4\u8fdb\u884c\u64cd\u4f5c\u3002\u5b83\u8fd8\u4f1a\u5c06\u7ef4\u5ea6\u6241\u5e73\u5316\u4e3a 1-D\uff0c\u8fd9\u662f\u8bb8\u591a\u540e\u7eed\u64cd\u4f5c\u6240\u9700\u7684\u683c\u5f0f\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> numpy as<\/span> np<\/span>\nfrom<\/span> openai.embeddings_utils import<\/span> distances_from_embeddings<\/span>\n<\/span>\ndf=pd.read_csv('processed\/embeddings.csv'<\/span>, index_col=0<\/span>)<\/span>\ndf['embeddings'<\/span>] = df['embeddings'<\/span>].apply(eval<\/span>).apply(np.array)<\/span>\n<\/span>\ndf.head()<\/span>\n<\/code><\/pre>\n
\u73b0\u5728\u6570\u636e\u5df2\u7ecf\u51c6\u5907\u597d\uff0c\u9700\u8981\u5c06\u95ee\u9898\u8f6c\u6362\u4e3a\u5177\u6709\u7b80\u5355\u51fd\u6570\u7684\u5d4c\u5165\u3002\u8fd9\u5f88\u91cd\u8981\uff0c\u56e0\u4e3a\u5d4c\u5165\u641c\u7d22\u4f7f\u7528\u4f59\u5f26\u8ddd\u79bb\u6bd4\u8f83\u6570\u5b57\u5411\u91cf\uff08\u8fd9\u662f\u539f\u59cb\u6587\u672c\u7684\u8f6c\u6362\uff09\u3002\u8fd9\u4e9b\u5411\u91cf\u53ef\u80fd\u662f\u76f8\u5173\u7684\uff0c\u5982\u679c\u5b83\u4eec\u7684\u4f59\u5f26\u8ddd\u79bb\u63a5\u8fd1\uff0c\u5219\u53ef\u80fd\u662f\u95ee\u9898\u7684\u7b54\u6848\u3002 OpenAI python \u5305\u6709\u4e00\u4e2a\u5185\u7f6e\u7684 distances_from_embeddings \u51fd\u6570\uff0c\u5728\u8fd9\u91cc\u5f88\u6709\u7528\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> create_context<\/span>(<\/span><\/span>\n    question, df, max_len=1800<\/span>, size=\"ada\"<\/span><\/span>\n):<\/span>\n    \"\"\"<\/span><\/span>\n    \u901a\u8fc7\u67e5\u627e\u6570\u636e\u6846\u4e2d\u6700\u76f8\u4f3c\u7684\u4e0a\u4e0b\u6587\u4e3a\u95ee\u9898\u521b\u5efa\u4e0a\u4e0b\u6587<\/span>\n    \"\"\"<\/span>\n<\/span>\n    # \u83b7\u53d6\u95ee\u9898\u7684\u5d4c\u5165\u5411\u91cf<\/span><\/span>\n    q_embeddings = client.embeddings.create(input<\/span>=question, engine='text-embedding-ada-002'<\/span>)['data'<\/span>][0<\/span>]['embedding'<\/span>]<\/span>\n<\/span>\n    # \u83b7\u53d6\u5d4c\u5165\u5411\u91cf\u4e4b\u95f4\u7684\u8ddd\u79bb<\/span><\/span>\n    df['distances'<\/span>] = distances_from_embeddings(q_embeddings, df['embeddings'<\/span>].values, distance_metric='cosine'<\/span>)<\/span>\n<\/span>\n<\/span>\n    returns = []<\/span>\n    cur_len = 0<\/span><\/span>\n<\/span>\n    # \u6309\u8ddd\u79bb\u6392\u5e8f\uff0c\u5e76\u6dfb\u52a0\u6587\u672c\u5230\u4e0a\u4e0b\u6587\u4e2d\uff0c\u76f4\u5230\u4e0a\u4e0b\u6587\u592a\u957f\u4e3a\u6b62<\/span><\/span>\n    for<\/span> i, row in<\/span> df.sort_values('distances'<\/span>, ascending=True<\/span>).iterrows():<\/span>\n<\/span>\n        # \u5c06\u6587\u672c\u7684\u957f\u5ea6\u6dfb\u52a0\u5230\u5f53\u524d\u957f\u5ea6<\/span><\/span>\n        cur_len += row['n_tokens'<\/span>] + 4<\/span><\/span>\n<\/span>\n        # \u5982\u679c\u4e0a\u4e0b\u6587\u592a\u957f\uff0c\u5219\u4e2d\u65ad<\/span><\/span>\n        if<\/span> cur_len > max_len:<\/span>\n            break<\/span><\/span>\n<\/span>\n        # \u5426\u5219\uff0c\u5c06\u5176\u6dfb\u52a0\u5230\u6b63\u5728\u8fd4\u56de\u7684\u6587\u672c\u4e2d<\/span><\/span>\n        returns.append(row[\"text\"<\/span>])<\/span>\n<\/span>\n    # \u8fd4\u56de\u4e0a\u4e0b\u6587<\/span><\/span>\n    return<\/span> \"nn###nn\"<\/span>.join(returns)<\/span>\n<\/span>\n<\/code><\/pre>\n
\u6587\u672c\u88ab\u5206\u89e3\u6210\u8f83\u5c0f\u7684\u6807\u8bb0\u96c6\uff0c\u56e0\u6b64\u6309\u5347\u5e8f\u5faa\u73af\u5e76\u7ee7\u7eed\u6dfb\u52a0\u6587\u672c\u662f\u786e\u4fdd\u5b8c\u6574\u7b54\u6848\u7684\u5173\u952e\u6b65\u9aa4\u3002\u5982\u679c\u8fd4\u56de\u7684\u5185\u5bb9\u8d85\u51fa\u9884\u671f\uff0c\u4e5f\u53ef\u4ee5\u5c06 max_len \u4fee\u6539\u5f97\u66f4\u5c0f\u3002<\/p>\n
\u4e0a\u4e00\u6b65\u53ea\u68c0\u7d22\u4e86\u4e0e\u95ee\u9898\u8bed\u4e49\u76f8\u5173\u7684\u6587\u672c\u5757\uff0c\u56e0\u6b64\u5b83\u4eec\u53ef\u80fd\u5305\u542b\u7b54\u6848\uff0c\u4f46\u4e0d\u80fd\u4fdd\u8bc1\u4e00\u5b9a\u5305\u542b\u7b54\u6848\u3002\u901a\u8fc7\u8fd4\u56de\u6700\u6709\u53ef\u80fd\u7684\u524d 5 \u4e2a\u7ed3\u679c\uff0c\u53ef\u4ee5\u8fdb\u4e00\u6b65\u63d0\u9ad8\u627e\u5230\u7b54\u6848\u7684\u51e0\u7387\u3002<\/p>\n
\u7136\u540e\uff0c\u7b54\u9898\u63d0\u793a\u4f1a\u5c1d\u8bd5\u4ece\u68c0\u7d22\u5230\u7684\u4e0a\u4e0b\u6587\u4e2d\u63d0\u53d6\u76f8\u5173\u4e8b\u5b9e\uff0c\u4ee5\u5f62\u6210\u4e00\u4e2a\u8fde\u8d2f\u7684\u7b54\u6848\u3002\u5982\u679c\u6ca1\u6709\u76f8\u5173\u7b54\u6848\uff0c\u63d0\u793a\u7b26\u5c06\u8fd4\u56de “\u6211\u4e0d\u77e5\u9053”\u3002<\/p>\n
\u53ef\u4ee5\u901a\u8fc7\u4f7f\u7528 gpt-3.5-turbo-instruct \u7684\u5b8c\u6210\u7ec8\u70b9\u6765\u521b\u5efa\u4e00\u4e2a\u542c\u8d77\u6765\u903c\u771f\u7684\u95ee\u9898\u7b54\u6848\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> answer_question<\/span>(<\/span><\/span>\n    df,<\/span>\n    model=\"gpt-3.5-turbo\"<\/span>,<\/span>\n    question=\"\u6211\u53ef\u4ee5\u5728\u4e0d\u7ecf\u4eba\u5de5\u5ba1\u6838\u7684\u60c5\u51b5\u4e0b\u5c06\u6a21\u578b\u8f93\u51fa\u53d1\u5e03\u5230Twitter\u5417\uff1f\"<\/span>,<\/span>\n    max_len=1800<\/span>,<\/span>\n    size=\"ada\"<\/span>,<\/span>\n    debug=False<\/span>,<\/span>\n    max_tokens=150<\/span>,<\/span>\n    stop_sequence=None<\/span><\/span>\n):<\/span>\n    \"\"\"<\/span><\/span>\n    \u57fa\u4e8e\u6570\u636e\u6846\u6587\u672c\u4e2d\u6700\u76f8\u4f3c\u7684\u4e0a\u4e0b\u6587\u56de\u7b54\u95ee\u9898<\/span>\n    \"\"\"<\/span>\n    context = create_context(<\/span>\n        question,<\/span>\n        df,<\/span>\n        max_len=max_len,<\/span>\n        size=size,<\/span>\n    )<\/span>\n    # \u5982\u679c\u8c03\u8bd5\uff0c\u6253\u5370\u539f\u59cb\u6a21\u578b\u54cd\u5e94<\/span><\/span>\n    if<\/span> debug:<\/span>\n        print<\/span>(\"\u4e0a\u4e0b\u6587:n\"<\/span> + context)<\/span>\n        print<\/span>(\"nn\"<\/span>)<\/span>\n<\/span>\n    try<\/span>:<\/span>\n        # \u4f7f\u7528\u95ee\u9898\u548c\u4e0a\u4e0b\u6587\u521b\u5efa\u804a\u5929\u5b8c\u6210<\/span><\/span>\n        response = client.chat.completions.create(<\/span>\n            model=\"gpt-3.5-turbo\"<\/span>,<\/span>\n            messages=[<\/span>\n                {\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: \"\u57fa\u4e8e\u4e0b\u9762\u7684\u4e0a\u4e0b\u6587\u56de\u7b54\u95ee\u9898\uff0c\u5982\u679c\u95ee\u9898\u65e0\u6cd5\u57fa\u4e8e\u4e0a\u4e0b\u6587\u56de\u7b54\uff0c\u5c31\u8bf4\u201c\u6211\u4e0d\u77e5\u9053\u201dnn\"<\/span>},<\/span>\n                {\"role\"<\/span>: \"user\"<\/span>, f\"content\"<\/span>: \"\u4e0a\u4e0b\u6587: {context}nn---nn\u95ee\u9898: {question}n\u56de\u7b54:\"<\/span>}<\/span>\n            ],<\/span>\n            temperature=0<\/span>,<\/span>\n            max_tokens=max_tokens,<\/span>\n            top_p=1<\/span>,<\/span>\n            frequency_penalty=0<\/span>,<\/span>\n            presence_penalty=0<\/span>,<\/span>\n            stop=stop_sequence,<\/span>\n        )<\/span>\n        return<\/span> response.choices[0<\/span>].message.strip()<\/span>\n    except<\/span> Exception as<\/span> e:<\/span>\n        print<\/span>(e)<\/span>\n        return<\/span> \"\"<\/span><\/span>\n<\/span>\n<\/code><\/pre>\n
\u5b8c\u6210\u4e86\uff01\u4e00\u4e2a\u5d4c\u5165\u4e86 OpenAI \u7f51\u7ad9\u77e5\u8bc6\u7684 Q\/A \u7cfb\u7edf\u73b0\u5df2\u51c6\u5907\u5c31\u7eea\u3002\u53ef\u4ee5\u8fdb\u884c\u4e00\u4e9b\u5feb\u901f\u6d4b\u8bd5\u6765\u67e5\u770b\u8f93\u51fa\u7684\u8d28\u91cf\uff1a<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>answer_question(df, question=\"What day is it?\"<\/span>, debug=False<\/span>)<\/span>\n<\/span>\nanswer_question(df, question=\"What is our newest embeddings model?\"<\/span>)<\/span>\n<\/span>\nanswer_question(df, question=\"What is ChatGPT?\"<\/span>)<\/span>\n<\/code><\/pre>\n
\u8fd4\u56de\u7684\u7ed3\u679c\u53ef\u80fd\u662f\uff1a<\/p>\n
<\/div>
text<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>\"I don't know.\"<\/span>\n<\/span>\n'The newest embeddings model is text-embedding-ada-002.'<\/span>\n<\/span>\n'ChatGPT is a model trained to interact in a conversational way. It is able to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.'<\/span>\n<\/code><\/pre>\n
\u5982\u679c\u7cfb\u7edf\u65e0\u6cd5\u56de\u7b54\u9884\u671f\u7684\u95ee\u9898\uff0c\u5219\u503c\u5f97\u641c\u7d22\u539f\u59cb\u6587\u672c\u6587\u4ef6\uff0c\u4ee5\u67e5\u770b\u9884\u671f\u5df2\u77e5\u7684\u4fe1\u606f\u662f\u5426\u5b9e\u9645\u4e0a\u6700\u7ec8\u88ab\u5d4c\u5165\u3002\u6700\u521d\u5b8c\u6210\u7684\u722c\u7f51\u8fc7\u7a0b\u88ab\u8bbe\u7f6e\u4e3a\u8df3\u8fc7\u6240\u63d0\u4f9b\u7684\u539f\u59cb\u57df\u4e4b\u5916\u7684\u7ad9\u70b9\uff0c\u56e0\u6b64\u5982\u679c\u5b58\u5728\u5b50\u57df\u8bbe\u7f6e\uff0c\u5b83\u53ef\u80fd\u4e0d\u77e5\u9053\u8fd9\u4e9b\u4fe1\u606f\u3002<\/p>\n
\u76ee\u524d\uff0c\u6bcf\u6b21\u90fd\u4f1a\u4f20\u9012\u6570\u636e\u5e27\u6765\u56de\u7b54\u95ee\u9898\u3002\u5bf9\u4e8e\u66f4\u591a\u7684\u751f\u4ea7\u5de5\u4f5c\u6d41\u7a0b\uff0c\u5e94\u4f7f\u7528\u77e2\u91cf\u6570\u636e\u5e93\u89e3\u51b3\u65b9\u6848\uff0c\u800c\u4e0d\u662f\u5c06\u5d4c\u5165\u5b58\u50a8\u5728 CSV \u6587\u4ef6\u4e2d\uff0c\u4f46\u5f53\u524d\u7684\u65b9\u6cd5\u662f\u539f\u578b\u8bbe\u8ba1\u7684\u7edd\u4f73\u9009\u62e9\u3002<\/p>","protected":false},"excerpt":{"rendered":"
\u672c\u6587\u4ecb\u7ecd\u4e86\u5982\u4f55\u5229\u7528OpenAI Embedding\u6a21\u578b\u548cGPT\u6784\u5efa\u667a\u80fd\u95ee\u7b54\u7cfb\u7edf\uff0c\u5305\u62ec\u722c\u53d6\u3001\u5d4c\u5165\u3001\u7d22\u5f15\u548c\u5b9e\u73b0\u8fc7\u7a0b\u3002<\/p>\n","protected":false},"author":1,"featured_media":11727,"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-1764","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\/1764","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=1764"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1764\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/11727"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1764"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1764"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1764"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1764"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}