使用附加提示进行解析
解析提示词 允许您以指导大型语言模型的相同方式,来引导我们的解析模型。
这些提示可用于提升解析器在复杂文档布局上的性能、以特定格式提取数据,或以其他方式转换文档。
在这个示例中,我们展示了如何通过向LlamaParse提供额外指令(提示)来塑造大型语言模型从非结构化文档中解析信息的方式。 使用麦当劳收据作为示例,我们演示了如何忽略文档的某些部分,仅解析每个订单的价格以及最终应付金额。
首先,我们设置环境并连接到LlamaCloud:
!pip install llama-cloud-servicesimport osfrom getpass import getpass
os.environ["LLAMA_CLOUD_API_KEY"] = getpass("Llama Cloud API Key: ")对于这个示例,我们使用以下麦当劳收据。请下载并将其保存为 mcdonalds_receipt.png:

我们首先初始化 LlamaParse,不包含任何指令:
from llama_cloud_services import LlamaParse
parser = LlamaParse( parse_mode="parse_page_with_agent", model="openai-gpt-4-1-mini", high_res_ocr=True, outlined_table_extraction=True, output_tables_as_HTML=True,)我们得到的结果如下:
vanilla_result = await parser.aparse("mcdonalds_receipt.png")print(vanilla_result.pages[0].md)Started parsing the file under job_id fa0c25a4-999f-439a-81a2-54f024ce2809
> Rate us HIGHLY SATISFIED and> Receive ONE FREE ITEM> Purchase any sandwich and receive an item of equal or lesser value> Go to www.mcdvoice.com within 7 days and tell us about your visit.> Validation Code:> Expires 30 days after receipt date.> Valid at participating US McDonald's.> Survey Code:> 31278-01121-21018-20481-00081-0
## McDonald's Restaurant #312782378 PINE RD NWRICE, MN 56367-9740TEL# 320 393 4600
| KS# 1 | 12/08/2022 08:48 PM ||-----------------|---------------------|| Side1 | Order 12 |
| Item | Price ||--------------------------|-------|| 1 Happy Meal 6 Pc | 4.89 || - 1 Creamy Ranch Cup | || - 1 Extra Kids Fry | || - 1 Wreck It Ralph 2 | || - 1 S Coke | || 1 Snack Oreo McFlurry | 2.69 |
| Subtotal | 7.58 || Tax | 0.52 || Take-Out Total | 8.10 |
| Cash Tendered | 10.00 || Change | 1.90 |
> McDonalds Restaurant Rice> ***NOW ACCEPTING APPLICATIONS***> text to #36453> apply31278现在让我们看看通过提供额外提示词来改变输出的方式:
parsing_instruction = """The provided document is a McDonald's receipt. Provide ONLY each line item (item name and price) and the final amount to be paid."""
parser = LlamaParse( parse_mode="parse_page_with_agent", model="openai-gpt-4-1-mini", high_res_ocr=True, outlined_table_extraction=True, output_tables_as_HTML=True, # Inject the parsing instruction into the user prompt user_prompt=parsing_instruction,)
result_with_prompt = await parser.aparse("mcdonalds_receipt.png")
print(result_with_prompt.pages[0].md)结果如下:
Started parsing the file under job_id 4c6a6443-0590-4384-b84d-65e4455f5e48
* Happy Meal 6 Pc 4.89* Snack Oreo McFlurry 2.69
Take-Out Total 8.10