\u8fd9\u4e9b\u65b9\u6848\u5728\u51c6\u786e\u6027\u3001\u53ef\u6269\u5c55\u6027\u548c\u6210\u672c\u6548\u76ca\u4e09\u65b9\u9762\u5f88\u96be\u627e\u5230\u5e73\u8861\u70b9\uff0c\u5c24\u5176\u5f53\u9762\u5bf9\u9700\u8981\u5904\u7406\u4e0a\u4ebf\u9875\u6587\u6863\u7684\u573a\u666f\u65f6\uff0c\u6210\u672c\u5f80\u5f80\u4ee4\u4eba\u671b\u800c\u5374\u6b65\u3002<\/p>\n\n\n\n \u8981\u5f00\u59cb\u4f7f\u7528Gemini 2.0\u5904\u7406PDF\u6587\u6863\uff0c\u9996\u5148\u9700\u8981\u8bbe\u7f6e\u73af\u5883\u5e76\u521b\u5efa\u63a8\u7406\u5ba2\u6237\u7aef\u3002\u4e0b\u9762\u662f\u5177\u4f53\u6b65\u9aa4\uff1a<\/p>\n\n\n\n \u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u6211\u4eec\u5df2\u7ecf\u5b8c\u6210\u4e86\u57fa\u7840\u73af\u5883\u914d\u7f6e\uff0c\u5e76\u6210\u529f\u4e0a\u4f20\u4e86\u7b2c\u4e00\u4e2aPDF\u6587\u4ef6\u7528\u4e8e\u5904\u7406\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cGemini\u6587\u4ef6API\u5141\u8bb8\u6bcf\u4e2a\u9879\u76ee\u6700\u591a\u5b58\u50a820GB\u7684\u6587\u4ef6\uff0c\u6bcf\u4e2a\u6587\u4ef6\u6700\u5927\u4e3a2GB\uff0c\u4e0a\u4f20\u7684\u6587\u4ef6\u4f1a\u4fdd\u5b5848\u5c0f\u65f6\u3002<\/p>\n\n\n\n Gemini 2.0\u7684\u4e00\u4e2a\u5f3a\u5927\u529f\u80fd\u662f\u80fd\u591f\u4ecePDF\u6587\u4ef6\u4e2d\u63d0\u53d6\u7ed3\u6784\u5316\u6570\u636e\u3002\u4e0b\u9762\u6211\u4eec\u5c06\u901a\u8fc7\u5b9e\u9645\u6848\u4f8b\u5c55\u793a\u5982\u4f55\u5229\u7528Pydantic\u6a21\u578b\u914d\u5408Gemini\u5b9e\u73b0\u8fd9\u4e00\u529f\u80fd\u3002<\/p>\n\n\n\n \u9996\u5148\uff0c\u6211\u4eec\u5b9a\u4e49\u4e00\u4e2a\u901a\u7528\u7684\u65b9\u6cd5\u6765\u5904\u7406PDF\u6587\u4ef6\u5e76\u8fd4\u56de\u7ed3\u6784\u5316\u6570\u636e\uff1a<\/p>\n\n\n\n \u5bf9\u4e8e\u53d1\u7968\u7c7bPDF\uff0c\u6211\u4eec\u53ef\u4ee5\u5b9a\u4e49\u4ee5\u4e0b\u6a21\u578b\u6765\u63d0\u53d6\u5173\u952e\u4fe1\u606f\uff1a<\/p>\n\n\n\n \u5bf9\u4e8e\u5305\u542b\u624b\u5199\u5185\u5bb9\u7684\u8868\u5355\uff0c\u6211\u4eec\u540c\u6837\u53ef\u4ee5\u5b9a\u4e49\u4e13\u7528\u7684\u6a21\u578b\uff1a<\/p>\n\n\n\n \u901a\u8fc7\u4e0a\u8ff0\u793a\u4f8b\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u5230Gemini 2.0\u80fd\u591f\u51c6\u786e\u8bc6\u522bPDF\u4e2d\u7684\u6587\u672c\u5185\u5bb9\uff0c\u751a\u81f3\u5305\u62ec\u624b\u5199\u6587\u5b57\uff0c\u5e76\u5c06\u5176\u8f6c\u6362\u4e3a\u7ed3\u6784\u5316\u7684JSON\u6570\u636e\u683c\u5f0f\uff0c\u6781\u5927\u5730\u7b80\u5316\u4e86\u6570\u636e\u63d0\u53d6\u6d41\u7a0b\u3002<\/p>\n\n\n\n \u5728RAG\uff08\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff09\u7cfb\u7edf\u4e2d\uff0c\u9664\u4e86\u57fa\u672c\u7684\u6587\u672c\u63d0\u53d6\u5916\uff0c\u6587\u6863\u5206\u5757\uff08Chunking\uff09\u4e5f\u662f\u5173\u952e\u6b65\u9aa4\u3002Gemini 2.0\u5141\u8bb8\u6211\u4eec\u5728\u4e00\u4e2a\u6b65\u9aa4\u4e2d\u5b8c\u6210OCR\u548c\u8bed\u4e49\u5206\u5757\u3002<\/p>\n\n\n\n \u4ee5\u4e0b\u662f\u4e00\u4e2a\u5c06PDF\u8f6c\u6362\u4e3aMarkdown\u5e76\u540c\u65f6\u8fdb\u884c\u8bed\u4e49\u5206\u5757\u7684\u63d0\u793a\u8bcd\uff1a<\/p>\n\n\n\n \u8fd9\u79cd\u65b9\u6cd5\u80fd\u591f\u8bc6\u522b\u6587\u6863\u7684\u8bed\u4e49\u8fb9\u754c\uff0c\u751f\u6210\u66f4\u6709\u610f\u4e49\u7684\u6587\u672c\u5757\uff0c\u6781\u5927\u5730\u63d0\u5347\u540e\u7eed\u68c0\u7d22\u7684\u51c6\u786e\u6027\u3002\u4e0e\u4f20\u7edf\u7684\u57fa\u4e8e\u5b57\u7b26\u6570\u7684\u673a\u68b0\u5206\u5757\u76f8\u6bd4\uff0c\u8bed\u4e49\u5206\u5757\u66f4\u80fd\u4fdd\u6301\u5185\u5bb9\u7684\u8fde\u8d2f\u6027\u548c\u5b8c\u6574\u6027\u3002<\/p>\n\n\n\n \u5bf9\u4e8e\u66f4\u590d\u6742\u7684\u573a\u666f\uff0c\u6211\u4eec\u53ef\u4ee5\u5b9a\u4e49\u5d4c\u5957\u7684Pydantic\u6a21\u578b\u6765\u5904\u7406\u591a\u5c42\u6b21\u7684\u6570\u636e\uff1a<\/p>\n\n\n\n \u5728\u5927\u89c4\u6a21\u5904\u7406PDF\u6587\u6863\u65f6\uff0c\u4ee5\u4e0b\u662f\u4e00\u4e9b\u63d0\u5347\u6548\u7387\u548c\u51c6\u786e\u5ea6\u7684\u6700\u4f73\u5b9e\u8df5\uff1a<\/p>\n\n\n\n \u5bf9\u4e8e\u9700\u8981\u5904\u7406\u5927\u91cfPDF\u7684\u573a\u666f\uff0c\u53ef\u4ee5\u5b9e\u73b0\u6279\u91cf\u5904\u7406\u6765\u63d0\u9ad8\u6548\u7387\uff1a<\/p>\n\n\n\n \u6839\u636e\u5b9e\u9645\u9700\u6c42\u9009\u62e9\u5408\u9002\u7684\u6a21\u578b\u53d8\u4f53\u53ef\u4ee5\u5927\u5e45\u964d\u4f4e\u6210\u672c\uff1a<\/p>\n\n\n\n \u4ee5\u4e0b\u662f\u5bf9\u4e0d\u540c\u6a21\u578b\u7684\u5904\u7406\u6548\u7387\u6bd4\u8f83\uff1a<\/p>\n\n\n\n \u5728\u751f\u4ea7\u73af\u5883\u4e2d\uff0c\u5b9e\u73b0\u7a33\u5065\u7684\u9519\u8bef\u5904\u7406\u673a\u5236\u81f3\u5173\u91cd\u8981\uff1a<\/p>\n\n\n\n \u5bf9\u4e8e\u542b\u6709\u590d\u6742\u8868\u683c\u7684PDF\uff0c\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u63d0\u793a\u8bcd\u6765\u63d0\u9ad8\u8868\u683c\u8bc6\u522b\u51c6\u786e\u7387\uff1a<\/p>\n\n\n\n![\"\"](\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/Jietu20250213-233957@2x.jpg\")
<\/figure>\n\n\n\n\u914d\u7f6e\u73af\u5883\u4e0e\u8bbe\u7f6eGemini 2.0<\/h2>\n\n\n\n
\u5b89\u88c5\u5fc5\u8981\u7684\u5e93<\/h3>\n\n\n\n
%pip install \"google-genai>=1\"\n<\/code><\/pre><\/div>\n\n\n\n\u521b\u5efa\u5ba2\u6237\u7aef\u4e0e\u6a21\u578b\u914d\u7f6e<\/h3>\n\n\n\n
from google import genai\n\n# \u521b\u5efa\u5ba2\u6237\u7aef\napi_key = \"YOUR_API_KEY\" # \u8bf7\u66ff\u6362\u4e3a\u4f60\u7684API\u5bc6\u94a5\nclient = genai.Client(api_key=api_key)\n\n# \u5b9a\u4e49\u8981\u4f7f\u7528\u7684\u6a21\u578b\nmodel_id = \"gemini-2.0-flash\" # \u4e5f\u53ef\u4ee5\u4f7f\u7528 \"gemini-2.0-flash-lite-preview-02-05\" \u6216 \"gemini-2.0-pro-exp-02-05\"\n<\/code><\/pre><\/div>\n\n\n\n\u4e0a\u4f20\u548c\u5904\u7406PDF\u6587\u4ef6<\/h3>\n\n\n\n
# \u4e0a\u4f20PDF\u6587\u4ef6\ninvoice_pdf = client.files.upload(file=\"invoice.pdf\", config={'display_name': 'invoice'})\n\n# \u67e5\u770b\u6587\u4ef6\u8f6c\u6362\u4e3a\u591a\u5c11\u4e2a\u4ee4\u724c\nfile_size = client.models.count_tokens(model=model_id, contents=invoice_pdf)\nprint(f'File: {invoice_pdf.display_name} equals to {file_size.total_tokens} tokens')\n\n# \u8f93\u51fa\u793a\u4f8b: File: invoice equals to 821 tokens\n<\/code><\/pre><\/div>\n\n\n\n\u7ed3\u6784\u5316PDF\u6570\u636e\u63d0\u53d6\u5b9e\u6218<\/h2>\n\n\n\n
\u5b9a\u4e49\u901a\u7528\u6570\u636e\u63d0\u53d6\u65b9\u6cd5<\/h3>\n\n\n\n
def extract_structured_data(file_path: str, model: BaseModel):\n # \u4e0a\u4f20\u6587\u4ef6\u5230\u6587\u4ef6API\n file = client.files.upload(file=file_path, config={'display_name': file_path.split('\/')[-1].split('.')[0]})\n \n # \u4f7f\u7528Gemini API\u751f\u6210\u7ed3\u6784\u5316\u54cd\u5e94\n prompt = f\"Extract the structured data from the following PDF file\"\n response = client.models.generate_content(model=model_id, \n contents=[prompt, file], \n config={'response_mime_type': 'application\/json', \n 'response_schema': model})\n \n # \u5c06\u54cd\u5e94\u8f6c\u6362\u4e3aPydantic\u6a21\u578b\u5e76\u8fd4\u56de\n return response.parsed\n<\/code><\/pre><\/div>\n\n\n\n\u6848\u4f8b1\uff1a\u53d1\u7968\u6570\u636e\u63d0\u53d6<\/h3>\n\n\n\n
from pydantic import BaseModel, Field\n\nclass Item(BaseModel):\n description: str = Field(description=\"The description of the item\")\n quantity: float = Field(description=\"The Qty of the item\")\n gross_worth: float = Field(description=\"The gross worth of the item\")\n\nclass Invoice(BaseModel):\n \"\"\"Extract the invoice number, date and all list items with description, quantity and gross worth and the total gross worth.\"\"\"\n invoice_number: str = Field(description=\"The invoice number e.g. 1234567890\")\n date: str = Field(description=\"The date of the invoice e.g. 2024-01-01\")\n items: list[Item] = Field(description=\"The list of items with description, quantity and gross worth\")\n total_gross_worth: float = Field(description=\"The total gross worth of the invoice\")\n\n# \u4f7f\u7528\u8be5\u6a21\u578b\u63d0\u53d6\u6570\u636e\nresult = extract_structured_data(\"invoice.pdf\", Invoice)\n\n# \u8f93\u51fa\u7ed3\u679c\nprint(f\"Extracted Invoice: {result.invoice_number} on {result.date} with total gross worth {result.total_gross_worth}\")\nfor item in result.items:\n print(f\"Item: {item.description} with quantity {item.quantity} and gross worth {item.gross_worth}\")\n<\/code><\/pre><\/div>\n\n\n\n![\"\"](\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/image-54.png\")
<\/figure>\n\n\n\n\u6848\u4f8b2\uff1a\u542b\u624b\u5199\u5185\u5bb9\u7684\u8868\u5355\u5904\u7406<\/h3>\n\n\n\n
class Form(BaseModel):\n \"\"\"Extract the form number, fiscal start date, fiscal end date, and the plan liabilities beginning of the year and end of the year.\"\"\"\n form_number: str = Field(description=\"The Form Number\")\n start_date: str = Field(description=\"Effective Date\")\n beginning_of_year: float = Field(description=\"The plan liabilities beginning of the year\")\n end_of_year: float = Field(description=\"The plan liabilities end of the year\")\n\n# \u63d0\u53d6\u6570\u636e\nresult = extract_structured_data(\"handwriting_form.pdf\", Form)\n\n# \u8f93\u51fa\u7ed3\u679c\nprint(f'Extracted Form Number: {result.form_number} with start date {result.start_date}.\\nPlan liabilities beginning of the year {result.beginning_of_year} and end of the year {result.end_of_year}')\n# \u8f93\u51fa\u793a\u4f8b: Extracted Form Number: CA530082 with start date 02\/05\/2022.\n# Plan liabilities beginning of the year 40000.0 and end of the year 55000.0\n<\/code><\/pre><\/div>\n\n\n\n\u9ad8\u7ea7\u5e94\u7528\uff1a\u6587\u6863\u5206\u5757\u4e0e\u8bed\u4e49\u7406\u89e3<\/h2>\n\n\n\n
PDF\u8bed\u4e49\u5206\u5757\u793a\u4f8b<\/h3>\n\n\n\n
CHUNKING_PROMPT = \"\"\"OCR the following page into Markdown. Tables should be formatted as HTML. \nDo not surround your output with triple backticks.\nChunk the document into sections of roughly 250 - 1000 words. Our goal is \nto identify parts of the page with same semantic theme. These chunks will \nbe embedded and used in a RAG pipeline. \nSurround the chunks with <chunk> <\/chunk> html tags.\"\"\"\n\n# \u4f7f\u7528\u8be5\u63d0\u793a\u8bcd\u8fdb\u884c\u5904\u7406\nresponse = client.models.generate_content(\n model=model_id,\n contents=[CHUNKING_PROMPT, pdf_file],\n)\n\nchunked_content = response.text\n<\/code><\/pre><\/div>\n\n\n\n\u4f7f\u7528Pydantic\u8fdb\u884c\u590d\u6742\u6570\u636e\u63d0\u53d6<\/h3>\n\n\n\n
class Person(BaseModel):\n first_name: str = Field(description=\"The first name of the person\")\n last_name: str = Field(description=\"The last name of the person\")\n age: int = Field(description=\"The age of the person, if not provided please return 0\")\n work_topics: list[Topic] = Field(description=\"The fields of interest of the person, if not provided please return an empty list\")\n\n# \u4f7f\u7528Person\u6a21\u578b\u751f\u6210\u54cd\u5e94\nprompt = \"Philipp Schmid is a Senior AI Developer Relations Engineer at Google DeepMind working on Gemini, Gemma with the mission to help every developer to build and benefit from AI in a responsible way.\"\nresponse = client.models.generate_content(\n model=model_id, \n contents=prompt, \n config={'response_mime_type': 'application\/json', 'response_schema': Person}\n)\n\n# SDK\u4f1a\u81ea\u52a8\u5c06\u54cd\u5e94\u8f6c\u6362\u4e3aPydantic\u6a21\u578b\nphilipp: Person = response.parsed\nprint(f\"First name is {philipp.first_name}\")\n<\/code><\/pre><\/div>\n\n\n\n\u6027\u80fd\u4f18\u5316\u4e0e\u6700\u4f73\u5b9e\u8df5<\/h2>\n\n\n\n
\u6279\u91cf\u5904\u7406\u4e0e\u4ee4\u724c\u4f18\u5316<\/h3>\n\n\n\n
async def batch_process_pdfs(file_paths, model, batch_size=10):\n results = []\n for i in range(0, len(file_paths), batch_size):\n batch = file_paths[i:i+batch_size]\n tasks = [extract_structured_data(path, model) for path in batch]\n batch_results = await asyncio.gather(*tasks)\n results.extend(batch_results)\n print(f\"Processed batch {i\/\/batch_size + 1}\/{(len(file_paths)+batch_size-1)\/\/batch_size}\")\n return results\n<\/code><\/pre><\/div>\n\n\n\n\u6a21\u578b\u9009\u62e9\u4e0e\u6210\u672c\u63a7\u5236<\/h3>\n\n\n\n
\n
\u6a21\u578b<\/strong><\/th> \u6bcf\u7f8e\u5143\u53ef\u5904\u7406\u7684PDF\u9875\u6570(Markdown\u8f6c\u6362)<\/strong><\/th><\/tr><\/thead> Gemini 2.0 Flash<\/td> \u7ea66,000\u9875<\/td><\/tr> Gemini 2.0 Flash Lite<\/td> \u7ea612,000\u9875<\/td><\/tr> Gemini 1.5 Flash<\/td> \u7ea610,000\u9875<\/td><\/tr> OpenAI 4o-mini<\/td> \u7ea6450\u9875<\/td><\/tr> OpenAI 4o<\/td> \u7ea6200\u9875<\/td><\/tr> Anthropic Claude-3.5<\/td> \u7ea6100\u9875<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n \u9519\u8bef\u5904\u7406\u4e0e\u91cd\u8bd5\u673a\u5236<\/h3>\n\n\n\n
def extract_with_retry(file_path, model, max_retries=3):\n for attempt in range(max_retries):\n try:\n return extract_structured_data(file_path, model)\n except Exception as e:\n if attempt == max_retries - 1:\n print(f\"Failed to process {file_path} after {max_retries} attempts: {e}\")\n return None\n print(f\"Attempt {attempt+1} failed, retrying: {e}\")\n time.sleep(2 ** attempt) # \u6307\u6570\u9000\u907f\u7b56\u7565\n<\/code><\/pre><\/div>\n\n\n\n![\"\"](\"https:\/\/school.myshirtai.com\/wp-content\/uploads\/2025\/05\/rd-bench-example.jpg\")
<\/figure>\n\n\n\n\u8868\u683c\u5904\u7406\u4f18\u5316<\/h3>\n\n\n\n