TruChain集成时出现Queue类型相关PyDantic验证错误求助

问题背景 #

我正在实现一个处理多选择题列表的process_questions函数，流程包括加载Embeddings模型、基于Redis搭建向量存储、创建处理链、遍历处理问题并记录结果。但在将处理链传入TruChain类时触发了Pydantic验证错误。

错误详情 #

self.__pydantic_validator__.validate_python(data, self_instance=self)
pydantic_core._pydantic_core.ValidationError: 1 validation error for TruChain
records_with_pending_feedback_results
  Input should be an instance of Queue [type=is_instance_of, input_value=<queue.Queue object at 0x124dd68b0>, input_type=Queue]

关联代码 #

from trulens_eval.feedback import Feedback, Groundedness
from trulens_eval import OpenAI as fOpenAI
from trulens_eval import TruChain
from trulens_eval.app import App
from typing import List

def process_questions(questions: List[QuestionType]):
    # Load the base embeddings model; specifics are abstracted
    embeddings = load_embeddings_model()
    # Initialize the vector store with the embeddings model
    vector_store = VectorStore(embeddings, "dimension_spec", chunk_size=1000, chunk_overlap=200)

    # Initialize the processing mechanism with the vector store's retriever
    processing_mechanism = ProcessingMechanism(vector_store.get_retriever())
    print("Processing Mechanism Type: ", type(processing_mechanism))

    # Construct and configure the processing chain
    processing_chain = processing_mechanism.construct_chain()
    processing_chain.bind(tags="ConfigurationTags")
    tru_recorder = TruChain(chain) ## 错误发生在此处

  
    # Iterate through the questions, process them, and handle responses
    counter = 1
    for question in questions:
        print(counter)
        counter += 1
        

        # Invoke the processing chain on the question data and record the process
        response, record = recorder.with_record(processing_chain.invoke, question.to_dict())
        print(f"Correct answer: {question.answer}")
        print('Response: ', response)

        # Output the recorded process in a structured format
        json_like = record.layout_as_app()
        print("Record Output: ", json_like)

修复/规避方案 #

1. 修正变量名错误 #

代码中TruChain(chain)里的chain变量未定义，这是明显的语法问题，先修正为实际创建的处理链变量：

tru_recorder = TruChain(processing_chain)

2. 升级trulens_eval版本 #

该错误多由版本兼容性问题导致，升级到最新稳定版通常能解决：

pip install --upgrade trulens_eval

3. 显式传入Queue实例 #

如果升级后问题仍存在，可手动创建并传入符合要求的Queue对象：

from queue import Queue

# ... 其他代码不变 ...

processing_chain = processing_mechanism.construct_chain()
processing_chain.bind(tags="ConfigurationTags")
tru_recorder = TruChain(processing_chain, records_with_pending_feedback_results=Queue())

4. 使用标准初始化流程 #

遵循TruChain官方推荐的初始化方式，先创建App实例再绑定到TruChain：

# ... 其他代码不变 ...

processing_chain = processing_mechanism.construct_chain()
processing_chain.bind(tags="ConfigurationTags")

app = App(processing_chain)
tru_recorder = TruChain(app)

内容的提问来源于stack exchange，提问作者JP1990