llm data_error ai_generated true

ValidationError: ResponseModel验证错误1个 name 字段必填 [type=missing, input_value={'title': 'Test'}, input_type=dict]

ValidationError: 1 validation error for ResponseModel name Field required [type=missing, input_value={'title': 'Test'}, input_type=dict]

ID: llm/langchain-output-parser-optional-field-missing

其他格式: JSON · Markdown 中文 · English
82%修复率
87%置信度
1证据数
2024-03-05首次发现

版本兼容性

版本状态引入弃用备注
langchain 0.1.0 active — — —
langchain 0.1.5 active — — —
langchain 0.2.0 active — — —
pydantic 2.5.0 active — — —
pydantic 2.6.0 active — — —

根因分析

LLM在JSON模式下的输出省略了Pydantic输出解析器模式中定义的必填字段，导致解析响应时验证失败。

English

LLM output in JSON mode omits a required field defined in the Pydantic output parser schema, causing validation failure when the response is parsed.

generic

官方文档

https://python.langchain.com/docs/modules/model_io/output_parsers/pydantic

解决方案

  1. 在系统提示中添加显式字段指令，确保LLM包含所有必填字段：

from langchain.output_parsers import PydanticOutputParser
parser = PydanticOutputParser(pydantic_object=ResponseModel)
prompt = PromptTemplate(
    template="生成JSON输出。确保以下字段始终存在：{format_instructions}\n{query}",
    input_variables=["query"],
    partial_variables={"format_instructions": parser.get_format_instructions()},
)
  2. 实现回退解析器，用None或默认值填充缺失字段，并记录遗漏以供监控：

try:
    parsed = parser.parse(llm_output)
except ValidationError:
    import json
    data = json.loads(llm_output)
    data.setdefault('name', 'unknown')
    parsed = ResponseModel(**data)

无效尝试

常见但无效的做法:

  1. 50% 失败

    Defeats the purpose of schema enforcement; LLM may skip critical fields entirely, leading to downstream data inconsistency.

  2. 70% 失败

    LLM behavior is non-deterministic; retrying the same prompt often yields the same omission pattern, especially with temperature=0.