LlamaIndex 中的结构化数据提取

使用 Pydantic 定义一个简单的模式，用于表示简历的顶层字段

1from pydantic import BaseModel
2from typing import List
3
4class ResumeData(BaseModel):
5    name: str
6    email: str
7    phone: str
8    skills: List[str]

使用 LlamaIndex 内置的 PDFReader 从 PDF 文件中加载简历文本

1from llama_index.readers.file import PDFReader
2
3pdf_reader = PDFReader()
4documents = pdf_reader.load_data("/path/to/resume.pdf")
5text = documents[0].text

连接到 Groq 托管的大语言模型，并利用模式提取结构化输出

1from llama_index.llms.groq import Groq
2
3llm = Groq(model="llama3-70b-8192", api_key="YOUR_GROQ_API_KEY")
4sllm = llm.as_structured_llm(ResumeData)
5response = sllm.complete(text)
6print(response)

使用基础模式时 LLM 返回的示例输出——仅包含简单的顶层字段

1ResumeData(
2    name="Priya Sharma",
3    email="priya.sharma@example.com",
4    phone="9876543210",
5    skills=["Python", "Java", "Docker"]
6)

使用 Pydantic 的 Field 为字段添加描述信息以增强模式定义

1from pydantic import Field
2
3class ResumeData(BaseModel):
4    name: str = Field(description="候选人的全名")
5    email: str = Field(description="候选人的电子邮件地址")
6    phone: str = Field(description="候选人的电话号码")
7    skills: List[str] = Field(description="候选人所具备的技术或专业技能列表")

调用结构化大模型并传入简历文本，使用带有字段描述的模式提取结构化数据

1sllm = llm.as_structured_llm(ResumeData)
2response = sllm.complete(text)

定义嵌套类以建模结构化的经历和教育条目

1from typing import Optional
2
3class ExperienceItem(BaseModel):
4    company_name: str = Field(description="候选人曾任职公司的名称")
5    job_title: str = Field(description="候选人在该公司担任的职位")
6    start_date: Optional[str] = Field(description="工作的开始日期")
7    end_date: Optional[str] = Field(description="工作的结束日期")
8
9class EducationItem(BaseModel):
10    degree: str = Field(description="学位或证书的名称")
11    institution: str = Field(description="大学或学校的名称")
12    graduation_year: Optional[int] = Field(description="候选人获得学位的年份")

更新主简历模式，加入嵌套的教育背景和工作经历字段

1class ResumeData(BaseModel):
2    name: str = Field(description="候选人的全名")
3    email: str = Field(description="候选人的电子邮件地址")
4    phone: str = Field(description="候选人的电话号码")
5    education: List[EducationItem] = Field(description="候选人教育经历的列表")
6    experience: List[ExperienceItem] = Field(description="候选人工作经历的列表")
7    skills: List[str] = Field(description="候选人所具备的技术或专业技能列表")

使用相同的 .as_structured_llm() 方法配合嵌套模式，获取更丰富的输出结果

1sllm = llm.as_structured_llm(ResumeData)
2response = sllm.complete(text)

使用嵌套模式时 LLM 返回的示例输出——包含工作经历与教育背景

1ResumeData(
2    name="Priya Sharma",
3    email="priya.sharma@example.com",
4    phone="9876543210",
5    education=[
6        EducationItem(degree="B.Tech in CS", institution="Delhi University", graduation_year=2017)
7    ],
8    experience=[
9        ExperienceItem(company_name="TechSpark", job_title="Backend Developer", start_date="2019-06-01", end_date="2022-08-30")
10    ],
11    skills=["Python", "Docker"]
12)

模式不匹配错误

1Error:
2openai.BadRequestError: Error code: 400 - {'error': {'message': "Failed to call a function. Please adjust your prompt. See 'failed_generation' for more details.", 'type': 'invalid_request_error', 'code': 'tool_use_failed', 'failed_generation': '<tool-use>{"tool_calls": [{"id": "pending", "type": "function", "function": {"name": "ResumeData"}, "parameters": {"name": "Alex Carter", "email": "alex.carter@example.com", "phone": "(+1) 555-123-4567", "education": [{"degree": "", "graduation_year": "", "institution": ""}], "experience": [{"company_name": "TechNova Institute", "job_title": "Software Engineer", "start_date": "Aug 2021", "end_date": "Present"}, {"company_name": "Global Digital University", "job_title": "Lecturer", "start_date": "Oct 2020", "end_date": "Feb 2020"}, {"company_name": "FutureWave Solutions Inc.", "job_title": "Senior Software Engineer & ML Specialist", "start_date": "Jun 2017", "end_date": "Sep 2020"}], "skills": ["AI", "machine learning", "enterprise software systems"]}}]}</tool-use>'}}
3Traceback:
4（此处省略具体堆栈跟踪）

修改模式以允许字段为可选字段

1class ExperienceItem(BaseModel):
2    company_name: Optional[str]
3    job_title: Optional[str]
4    start_date: Optional[str]
5    end_date: Optional[str]
6
7class EducationItem(BaseModel):
8    degree: Optional[str]
9    institution: Optional[str]
10    graduation_year: Optional[int]

通过设置 strict=False 放宽验证规则

1sllm = llm.as_structured_llm(ResumeData, strict=False)
2response = sllm.complete(text)

错误

1Failed to call a function. Please adjust your prompt.