{"id":1165,"date":"2024-04-22T15:33:14","date_gmt":"2024-04-22T07:33:14","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1165"},"modified":"2024-04-22T15:33:19","modified_gmt":"2024-04-22T07:33:19","slug":"zhinanopenaiapiduipdfjinxingzongjietilian","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/zhinanopenaiapiduipdfjinxingzongjietilian.html","title":{"rendered":"\u6307\u5357 | OpenAI API\u5bf9PDF\u8fdb\u884c\u603b\u7ed3\u63d0\u70bc"},"content":{"rendered":"

\u524d\u8a00<\/h2>\n

\u672c\u6587\u5c06\u9488\u5bf9\u4fe1\u606f\u8fc7\u8f7d\u9605\u8bfb\u641c\u7d22\u96be\u7684\u95ee\u9898\uff0c\u5c55\u793a\u5982\u4f55\u5229\u7528OpenAI API\u9ad8\u6548\u5730\u81ea\u52a8\u603b\u7ed3PDF\uff0c\u4ece\u800c\u63d0\u9ad8\u9605\u8bfb\u6548\u7387\u548c\u8282\u7701\u65f6\u95f4\u3002\u5305\u542b\u8be6\u5c3d\u7684\u64cd\u4f5c\u6559\u7a0b\u3001\u5b9e\u9645\u6848\u4f8b\u53ca\u6700\u4f73\u5b9e\u8df5\uff0c\u65e8\u5728\u786e\u4fdd\u51c6\u786e\u4e14\u9ad8\u6548\u5730\u8fdb\u884c\u6587\u6863\u6458\u8981\u3002\u53e6\u5916\uff0c\u672c\u6587\u8fd8\u5c06\u6df1\u5165\u63a2\u8ba8\u8fd0\u7528AI\u6280\u672f\u8fdb\u884cPDF\u603b\u7ed3\u7684\u4f18\u52bf\uff0c\u5982\u63d0\u9ad8\u751f\u4ea7\u529b\u3001\u8282\u7701\u65f6\u95f4\u7b49\uff0c\u540c\u65f6\u5206\u6790\u4e86\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u53ef\u80fd\u9047\u5230\u7684\u5c40\u9650\u6027\u548c\u6311\u6218\u3002<\/p>\n

\n

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

\u901a\u8fc7\u672c\u6587\uff0c\u53ef\u5e2e\u52a9\u60a8\u7b80\u5316\u9605\u8bfb\u8fc7\u7a0b\uff0c\u63d0\u9ad8\u5904\u7406\u5927\u91cf\u4fe1\u606f\u7684\u80fd\u529b\uff0c\u4ece\u800c\u63d0\u5347\u5de5\u4f5c\u6548\u7387\u548c\u5b66\u672f\u8868\u73b0\uff0c\u66f4\u597d\u5730\u5e94\u5bf9\u5feb\u8282\u594f\u4e16\u754c\u4e2d\u7684\u4fe1\u606f\u6311\u6218\uff0c\u5145\u5206\u5229\u7528AI\u6280\u672f\u4e3a\u60a8\u7684\u9605\u8bfb\u548c\u5b66\u4e60\u5e26\u6765\u66f4\u591a\u4fbf\u5229\u3002<\/p>\n

\u73af\u5883\u51c6\u5907<\/h2>\n

\u5728\u5f00\u59cb\u524d\uff0c\u6211\u4eec\u8981\u51c6\u5907\u597d\u5f00\u53d1\u5f00\u53d1\u73af\u5883\uff0c\u9996\u5148\u4f60\u8981\u51c6\u5907\u597dOpenai<\/code>\u7684\u8bbf\u95ee\u5bc6\u94a5\uff0c\u63d0\u4f9b\u6587\u672c\u603b\u7ed3\u7684\u80fd\u529b\u3002
\n\u5176\u6b21\u4f60\u8981\u51c6\u5907python<\/code>\u8bed\u8a00\u7684\u5f00\u53d1\u73af\u5883\uff0c\u5e76\u4e14\u5b89\u88c5Fitz<\/code>\u5e93\uff0cA\u8be5\u5e93\u5c06\u63d0\u4f9b\u8bfb\u53d6\u3001\u5199\u5165\u548c\u64cd\u4f5cPDF\u6587\u4ef6\u7684\u80fd\u529b\u3002<\/p>\n

<\/div>
pip<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>pip install PyMuPDF openai<\/span>\n<\/code><\/pre>\n
\u7136\u540e\u5c06\u5fc5\u8981\u7684\u5e93\u5f15\u5165\u5230Python\u6587\u4ef6\u4e2d\u3002<\/p>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> fitz<\/span>\nimport<\/span> openai<\/span>\n<\/code><\/pre>\n
\u8bfb\u53d6pdf\u5185\u5bb9<\/h2>\n
<\/div>
python<\/span>
\u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> fitz<\/span>\nimport<\/span> openai<\/span>\n<\/span>\ncontext = \"\"<\/span><\/span>\n<\/span>\nwith<\/span> fitz.open<\/span>(\u2018filename.pdf\u2019) as<\/span> pdf_file:<\/span>\nnum_pages = pdf_file.page_count<\/span>\n<\/span>\nfor<\/span> page_num in<\/span> range<\/span>(num_pages):<\/span>\n    page = pdf_file[page_num]<\/span>\n    page_text = page.get_text()<\/span>\n    context += page_text<\/span>\n<\/code><\/pre>\n
    \n
  1. \u521b\u5efa\u4e00\u4e2a\u540d\u4e3aContext\u7684\u53d8\u91cf\uff0c\u5c06\u5176\u8bbe\u7f6e\u4e3a\u7a7a\u5b57\u7b26\u4e32\u3002\u6b64\u53d8\u91cf\u5c06\u7528\u4e8e\u5b58\u50a8\u6574\u4e2aPDF\u6587\u6863\u7684\u6587\u672c\u5185\u5bb9\u3002<\/li>\n
  2. \u4f7f\u7528Fitz\u5e93\u7684fitz.open()\u51fd\u6570\u6253\u5f00PDF\u6587\u4ef6\u3002\u6b64\u51fd\u6570\u5c06PDF\u6587\u4ef6\u540d\u4f5c\u4e3a\u53c2\u6570\u3002<\/li>\n
  3. \u4f7f\u7528pdf_file\u5bf9\u8c61\u7684page_count\u5c5e\u6027\u83b7\u53d6PDF\u6587\u4ef6\u4e2d\u7684\u603b\u9875\u6570\u3002\u8fd9\u5c06\u7528\u4e8e\u5faa\u73af\u904d\u5386PDF\u6587\u4ef6\u4e2d\u7684\u6bcf\u4e00\u9875\u3002<\/li>\n
  4. \u4f7f\u7528for\u5faa\u73af\u904d\u5386PDF\u6587\u4ef6\u4e2d\u7684\u6bcf\u4e00\u9875\u3002\u5faa\u73af\u4ece0\u5f00\u59cb\uff0c\u5230num_pages – 1\u7ed3\u675f\uff0c\u56e0\u4e3aPython\u7684\u6570\u7ec4\u7d22\u5f15\u4ece0\u5f00\u59cb\u3002<\/li>\n
  5. \u5728\u5faa\u73af\u5185\u90e8\uff0c\u4f7f\u7528pdf_file\u5bf9\u8c61\u83b7\u53d6\u5f53\u524d\u9875\u9762\u3002page_num\u53d8\u91cf\u7528\u4e8e\u7d22\u5f15PDF\u6587\u4ef6\u4e2d\u7684\u9875\u9762\u3002<\/li>\n
  6. \u4f7f\u7528\u9875\u9762\u5bf9\u8c61\u7684get_text()\u65b9\u6cd5\u63d0\u53d6\u5f53\u524d\u9875\u9762\u7684\u6587\u672c\u5185\u5bb9\u3002\u6b64\u65b9\u6cd5\u8fd4\u56de\u5305\u542b\u9875\u9762\u6587\u672c\u7684\u5b57\u7b26\u4e32\u3002<\/li>\n
  7. \u4f7f\u7528+=\u8fd0\u7b97\u7b26\u5c06\u5f53\u524d\u9875\u9762\u7684\u6587\u672c\u8ffd\u52a0\u5230Context\u53d8\u91cf\u3002\u8fd9\u5c06\u6587\u672c\u6dfb\u52a0\u5230Context\u5b57\u7b26\u4e32\u7684\u672b\u5c3e\u3002<\/li>\n
  8. \u5728\u5faa\u73af\u5b8c\u6210\u540e\uff0cContext\u53d8\u91cf\u5c06\u5305\u542b\u6574\u4e2aPDF\u6587\u6863\u7684\u6587\u672c\u5185\u5bb9\u3002<\/li>\n<\/ol>\n
    \u5206\u89e3pdf\u6210\u6bb5\u843d<\/h2>\n
    \u4e0b\u4e00\u6b65\u662f\u5c06pdf\u7684\u5185\u5bb9\u8f93\u5165 split_text <\/code>\u51fd\u6570\uff0c\u8be5\u51fd\u6570\u8d1f\u8d23\u5c06\u6587\u672c\u5212\u5206\u4e3a\u6bcf\u6bb5\u5305\u542b5000\u4e2a\u5b57\u7b26\u7684\u6bb5\u843d\u3002<\/p>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div><\/span>\ndef<\/span> split_text<\/span>(text, chunk_size=5000<\/span><\/span>):<\/span>\n\u201c\u201d\u201d<\/span>\nSplits the given text into chunks of approximately the specified chunk size.<\/span>\n<\/span>\nArgs:<\/span>\ntext (str<\/span>): The text to split.<\/span>\n<\/span>\nchunk_size (int<\/span>): The desired size of each chunk (in<\/span> characters).<\/span>\n<\/span>\nReturns:<\/span>\nList<\/span>[str<\/span>]: A list<\/span> of chunks, each of approximately the specified chunk size.<\/span>\n\u201c\u201d\u201d<\/span>\n<\/span>\nchunks = []<\/span>\ncurrent_chunk = StringIO()<\/span>\ncurrent_size = 0<\/span><\/span>\nsentences = sent_tokenize(text)<\/span>\nfor<\/span> sentence in<\/span> sentences:<\/span>\n  sentence_size = len<\/span>(sentence)<\/span>\n  if<\/span> sentence_size > chunk_size:<\/span>\n    while<\/span> sentence_size > chunk_size:<\/span>\n      chunk = sentence[:chunk_size]<\/span>\n      chunks.append(chunk)<\/span>\n      sentence = sentence[chunk_size:]<\/span>\n      sentence_size -= chunk_size<\/span>\n      current_chunk = StringIO()<\/span>\n      current_size = 0<\/span><\/span>\n  if<\/span> current_size + sentence_size < chunk_size:<\/span>\n    current_chunk.write(sentence)<\/span>\n    current_size += sentence_size<\/span>\n  else<\/span>:<\/span>\n    chunks.append(current_chunk.getvalue())<\/span>\n    current_chunk = StringIO()<\/span>\n    current_chunk.write(sentence)<\/span>\n    current_size = sentence_size<\/span>\nif<\/span> current_chunk:<\/span>\n   chunks.append(current_chunk.getvalue())<\/span>\nreturn<\/span> chunks<\/span>\n<\/code><\/pre>\n
      \n
    1. \u8be5\u4ee3\u7801\u5b9a\u4e49\u4e86\u4e00\u4e2a\u540d\u4e3a split_text<\/code> \u7684 Python \u51fd\u6570\uff0c\u7528\u4e8e\u5c06\u8f83\u957f\u7684\u6587\u672c\u5206\u5272\u6210\u5927\u5c0f\u76f8\u8fd1\u7684\u8f83\u5c0f\u5757\u3002<\/li>\n
    2. \u51fd\u6570\u4f7f\u7528 sent_tokenize<\/code> \u51fd\u6570\u5c06\u6587\u672c\u5206\u5272\u6210\u53e5\u5b50\uff0c\u7136\u540e\u904d\u5386\u6bcf\u4e2a\u53e5\u5b50\u4ee5\u521b\u5efa\u5757\u3002<\/li>\n
    3. \u6bcf\u4e2a\u5757\u7684\u5927\u5c0f\u53ef\u4ee5\u5728\u51fd\u6570\u53c2\u6570\u4e2d\u6307\u5b9a\uff0c\u9ed8\u8ba4\u5927\u5c0f\u4e3a5000\u4e2a\u5b57\u7b26\u3002<\/li>\n
    4. \u8be5\u51fd\u6570\u8fd4\u56de\u4e00\u4e2a\u5757\u5217\u8868\uff0c\u6bcf\u4e2a\u5757\u7684\u5927\u5c0f\u7ea6\u4e3a\u6307\u5b9a\u7684\u5927\u5c0f\u3002\n\u8fd9\u4e2a\u51fd\u6570\u53ef\u4ee5\u7528\u4e8e\u5bf9\u8f83\u957f\u7684 PDF \u6587\u672c\u8fdb\u884c\u6458\u8981\u5904\u7406\uff0c\u5c06\u5b83\u4eec\u5206\u89e3\u6210\u5bb9\u6613\u5904\u7406\u7684\u8f83\u5c0f\u5757\u3002<\/li>\n<\/ol>\n
      \u603b\u7684\u6765\u8bf4\uff0csplit_text<\/code>\u51fd\u6570\u5c06\u5197\u957f\u7684PDF\u6587\u672c\u5206\u89e3\u6210\u53ef\u7ba1\u7406\u7684\u6bb5\u843d\uff0c\u4ee5\u4fbf\u8fdb\u884c\u5f52\u7eb3\u603b\u7ed3\uff0c\u4f7f\u5176\u66f4\u5bb9\u6613\u5904\u7406\u548c\u63d0\u53d6\u5173\u952e\u4fe1\u606f\u3002<\/p>\n
      \u7528OpenAI\u6a21\u578b\u5f52\u7eb3\u603b\u7ed3\u6587\u672c\u6bb5\u843d<\/h2>\n
      <\/div>
      python<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> gpt3_completion<\/span>(prompt, engine=\u2019text-davinci-003', temp=0.5, top_p=0.3, tokens=1000):<\/span><\/span><\/span>\n<\/span>\nprompt = prompt.encode(encoding=\u2019ASCII\u2019,errors=\u2019ignore\u2019).decode()<\/span>\ntry:<\/span>\n  response = openai.Completion.create(<\/span>\n  engine=engine,<\/span>\n  prompt=prompt,<\/span>\n  temperature=temp,<\/span>\n  top_p=top_p,<\/span>\n  max_tokens=tokens<\/span>\n  )<\/span>\n  return response.choices[0].text.strip()<\/span>\nexcept Exception as oops:<\/span>\n  return \u201cGPT-3 error: %s\u201d % oops<\/span>\n<\/span>\n<\/code><\/pre>\n
        \n
      1. def gpt3_completion(prompt, engine='text-davinci-003', temp=0.5, top_p=0.3, tokens=1000) <\/code>\u8fd9\u4e2a\u51fd\u6570\u63a5\u53d7\u4e00\u4e2a\u63d0\u793a\u5b57\u7b26\u4e32\u548c\u56db\u4e2a\u53ef\u9009\u53c2\u6570\uff1aengine\uff0ctemp\uff0ctop_p \u548c tokens\u3002<\/li>\n
      2. promptprompt.encode(encoding='ASCII',errors='ignore').decode()<\/code> \u5c06\u63d0\u793a\u5b57\u7b26\u4e32\u7f16\u7801\u4e3a ASCII \u683c\u5f0f\uff0c\u5e76\u5728\u7f16\u7801\u8fc7\u7a0b\u4e2d\u5ffd\u7565\u53ef\u80fd\u51fa\u73b0\u7684\u4efb\u4f55\u9519\u8bef\uff0c\u7136\u540e\u5c06\u5176\u89e3\u7801\u56de\u5b57\u7b26\u4e32\u3002\n3. try:<\/code> \u548cexcept Exception as oops:<\/code>\u7528\u4e8e\u5904\u7406\u5728\u6267\u884c try \u5757\u5185\u7684\u4ee3\u7801\u65f6\u53ef\u80fd\u51fa\u73b0\u7684\u4efb\u4f55\u9519\u8bef\u3002<\/li>\n
      3. response = openai.Completion.create(...) <\/code>\u5411 OpenAI GPT-3 API <\/code>\u53d1\u9001\u8bf7\u6c42\uff0c\u6839\u636e\u63d0\u4f9b\u7684prompt\u3001engine\u3001temp\u3001top_p \u548c tokens \u53c2\u6570\u751f\u6210\u6587\u672c\u8865\u5168\u3002\u54cd\u5e94\u7ed3\u679c\u5b58\u50a8\u5728 response \u53d8\u91cf\u4e2d\u3002<\/li>\n
      4. return response.choices[0].text.strip()<\/code> \u4ece\u54cd\u5e94\u5bf9\u8c61\u4e2d\u63d0\u53d6\u751f\u6210\u7684\u6587\u672c\uff0c\u5e76\u5220\u9664\u4efb\u4f55\u524d\u5bfc\u6216\u5c3e\u968f\u7684\u7a7a\u767d\u5b57\u7b26\u3002<\/li>\n
      5. \u5982\u679c\u5728\u6267\u884c try \u5757\u671f\u95f4\u53d1\u751f\u9519\u8bef\uff0c\u51fd\u6570\u5c06\u8fd4\u56de\u4e00\u4e2a\u5305\u542b\u9519\u8bef\u4fe1\u606f\u7684\u9519\u8bef\u6d88\u606f\u3002<\/li>\n<\/ol>\n
        \u603b\u7ed3<\/h2>\n
        \u603b\u7684\u6765\u8bf4\uff0c\u672c\u6587\u5b9e\u73b0\u7684\u80fd\u529b\u6781\u5927\u5730\u7b80\u5316\u9605\u8bfb\u8fc7\u7a0b\u3002\u901a\u8fc7\u4f7f\u7528 Fitz \u5e93\u548c OpenAI \u7684 GPT-3 \u8bed\u8a00\u6a21\u578b\u6211\u4eec\u53ef\u4ee5\u8f7b\u677e\u5730\u4ece PDF \u6587\u4ef6\u4e2d\u63d0\u53d6\u5e76\u603b\u7ed3\u5173\u952e\u4fe1\u606f\u3002\u53e6\u5916\uff0c\u672c\u6587\u4f7f\u7528\u7684\u662fGPT-3\u6a21\u578b\uff0c\u5982\u679c\u4f7f\u7528GPT-4\u6a21\u578b\u90a3\u6548\u679c\u66f4\u4f73\uff0c\u7279\u522b\u662f2023\u5e746.13\u4ee5\u540e\u7684\u6a21\u578b\u7248\u672c\u66f4\u51fa\u5f69\u3002<\/p>\n
        \n
        \"image.png\"<\/figure>\n<\/p>\n
        \u5f53\u7136\uff0c\u672c\u6587\u53ea\u662f\u4e3a\u5b9e\u73b0pdf\u7684\u603b\u7ed3\u63d0\u70bc\u63d0\u4f9b\u4e00\u4e2a\u601d\u8def\uff0c\u5982\u679c\u662f\u5de5\u7a0b\u7ea7\u522b\uff0c\u5176\u4e2d\u8fd8\u8981\u8003\u8651\u5230\u5904\u7406\u6587\u4ef6\u9875\u6570\u8db3\u591f\u7684\u591a\u600e\u4e48\u529e\uff0c\u5982\u4f55\u7d22\u5f15\u76ee\u5f55\u4ee5\u53ca\u5185\u5bb9\uff0c\u5982\u4f55\u907f\u514d\u6a21\u578b\u80e1\u8bcc\uff0c\u5982\u4f55\u63d0\u4f9b\u9690\u79c1\u52a0\u5bc6\u7b49\u7b49\u3002\u5176\u4e2d\u53ef\u80fd\u8981\u7528\u5230Embedding\uff0c\u63d0\u793a\u5de5\u7a0b\uff0c\u52a0\u5bc6\u7b49\u6280\u672f\uff0c\u53ef\u53c2\u7167\u4e0a\u56fe\u7684\u5927\u4f53\u6d41\u7a0b\uff0c\u6211\u4eec\u653e\u5728\u540e\u9762\u7684\u6587\u7ae0\u8ba8\u8bba\u3002<\/p>","protected":false},"excerpt":{"rendered":"
        \u603b\u7684\u6765\u8bf4\uff0c\u672c\u6587\u5b9e\u73b0\u7684\u80fd\u529b\u6781\u5927\u5730\u7b80\u5316\u9605\u8bfb\u8fc7\u7a0b\u3002\u901a\u8fc7\u4f7f\u7528 Fitz \u5e93\u548c OpenAI \u7684 GPT-3 \u8bed\u8a00\u6a21\u578b\u6211\u4eec\u53ef\u4ee5\u8f7b\u677e\u5730\u4ece PDF \u6587\u4ef6\u4e2d\u63d0\u53d6\u5e76\u603b\u7ed3\u5173\u952e\u4fe1\u606f\u3002<\/p>\n","protected":false},"author":1,"featured_media":1891,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"PDF | \u51fd\u6570 | sentencesize - \u5948\u65af\u79d1\u6280\u793e\u533a","rank_math_description":"\u5948\u65af\u79d1\u6280\u793e\u533a\u4e13\u6ce8\u4e8ePDF\u3001\u51fd\u6570\u3001sentencesize\u3001chunksize\u3001python\uff0c\u63d0\u4f9b\u4e13\u4e1a\u7684\u5206\u4eab\u548c\u8ba8\u8bba\u5e73\u53f0\u3002","rank_math_focus_keyword":"PDF,\u51fd\u6570,sentencesize,chunksize,python","views":"62","footnotes":""},"categories":[3],"tags":[126,127,128,129,136],"collection":[],"class_list":["post-1165","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\/1165","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=1165"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1165\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/1891"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1165"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1165"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1165"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1165"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}