如何修复Python中Ragas库的AttributeError：'str'无'generator_llm'属性

问题背景 #

同一份Ragas生成测试集的代码在Notebook中可正常运行，但转为脚本或在Notebook中导入脚本调用时，执行到100%后报错，报错信息显示'str' object has no attribute 'generator_llm'。脚本与Notebook使用的Ragas版本均为0.1.4。

原代码 #

import importlib.resources
import os

import pandas as pd
import yaml
from dotenv import load_dotenv
from google.cloud import storage
from google.cloud.storage import Blob
from langchain.docstore.document import Document
from langchain_community.document_loaders import PDFPlumberLoader
from langchain_experimental.text_splitter import SemanticChunker
from langchain_openai.embeddings import OpenAIEmbeddings
from ragas.testset.evolutions import multi_context, reasoning, simple
from ragas.testset.generator import TestsetGenerator
from tqdm.notebook import tqdm

load_dotenv()


def get_pdf_files(collection_name, filegroup_name):
    """Get the pdf files from the collection and filegroup from the ds-librairie-genai-provisioning"""

    collections_path = importlib.resources.files("genai_provisioning").joinpath(
        "collections"
    )
    collection_path = collections_path.joinpath(collection_name)

    filegroups_path = collection_path.joinpath("filegroups")
    files_directory = filegroups_path.joinpath(filegroup_name).joinpath("files")

    pdf_files = list(files_directory.glob("*.pdf"))

    return pdf_files


def combine_all_page(docs):
    """Combine all pages from one PDF into one Document object"""
    try:
        page_content = "/n".join([doc.page_content for doc in docs])
        page_metadata = docs[0].metadata
        if "page" in page_metadata:
            page_metadata.pop("page")
        return [Document(page_content=page_content, metadata=page_metadata)]
    except Exception as e:
        print(e)


def generate_testset(configs_file):
    """Generate RAGAS synthetic test set"""
    with open(configs_file) as f:
        configs = yaml.safe_load(f)

    collection_name = configs["collection_name"]
    filegroup_name = configs["filegroup_name"]
    version = configs["version"]

    files = get_pdf_files(collection_name, filegroup_name)
    print(f"Number of files: {len(files)}")

    docs = []
    for file in tqdm(files):
        doc_ = PDFPlumberLoader(
            str(file),
            text_kwargs={
                "x_tolerance": configs["x_tolerance"],
                "y_tolerance": configs["y_tolerance"],
            },
        ).load()
        doc_ = combine_all_page(doc_)
        docs.extend(doc_)
    print(f"Number of documents: {len(docs)}")

    for doc_ in docs:
        doc_.metadata["filename"] = doc_.metadata["source"]

    splitted_testset_docs = docs

    # RAGAS to generate a synthetic testset
    generator = TestsetGenerator.with_openai()

    testset = generator.generate_with_langchain_docs(
        splitted_testset_docs,
        test_size=configs["test_size"],
        with_debugging_logs=True,
        distributions=configs["distributions"],
    )

报错信息 #

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[2], line 3
      1 from generate_testset import generate_testset
----> 3 generate_testset("generate-testset.yml")

File ~/github_repos/ds-research-genai/common/src/rag_pipeline_evaluator/generate_testset.py:86, in generate_testset(configs_file)
     83 print("len(splitted_testset_docs)", len(splitted_testset_docs))
     84 print("splitted_testset_docs[0]", splitted_testset_docs[0])
---> 86 testset = generator.generate_with_langchain_docs(
     87     splitted_testset_docs,
     88     test_size=configs["test_size"],
     89     with_debugging_logs=True,
     90     distributions=configs["distributions"],
     91 )
     93 # Remove nan and short ground truth
     94 testset_improved = testset

File ~/github_repos/ds-research-genai/common/src/rag_pipeline_evaluator/venv/lib/python3.11/site-packages/ragas/testset/generator.py:179, in TestsetGenerator.generate_with_langchain_docs(self, documents, test_size, distributions, with_debugging_logs, is_async, raise_exceptions, run_config)
    174 # chunk documents and add to docstore
    175 self.docstore.add_documents(
    176     [Document.from_langchain_document(doc) for doc in documents]
    177 )
--> 179 return self.generate(
    180     test_size=test_size,
    181     distributions=distributions,
    182     with_debugging_logs=with_debugging_logs,
    183     is_async=is_async,
    184     raise_exceptions=raise_exceptions,
    185     run_config=run_config,
    186 )

File ~/github_repos/ds-research-genai/common/src/rag_pipeline_evaluator/venv/lib/python3.11/site-packages/ragas/testset/generator.py:227, in TestsetGenerator.generate(self, test_size, distributions, with_debugging_logs, is_async, raise_exceptions, run_config)
    225 # init filters and evolutions
    226 for evolution in distributions:
--> 227     self.init_evolution(evolution)
    228     evolution.init(is_async=is_async, run_config=run_config)
    230 if with_debugging_logs:

File ~/github_repos/ds-research-genai/common/src/rag_pipeline_evaluator/venv/lib/python3.11/site-packages/ragas/testset/generator.py:189, in TestsetGenerator.init_evolution(self, evolution)
    188 def init_evolution(self, evolution: Evolution) -> None:
--> 189     if evolution.generator_llm is None:
    190         evolution.generator_llm = self.generator_llm
    191         if evolution.docstore is None:

AttributeError: 'str' object has no attribute 'generator_llm'

问题原因 #

1. distributions类型不匹配：YAML配置文件中distributions存储的是字符串（如["simple", "reasoning"]），但Ragas的generate_with_langchain_docs方法需要传入的是从ragas.testset.evolutions导入的Evolution类实例。Notebook中可能直接使用了对象而非从YAML读取字符串，因此未触发该问题。
2. tqdm导入错误：脚本中使用from tqdm.notebook import tqdm，该模块仅适用于Notebook环境，脚本运行时会引发界面相关错误。

修复方案 #

1. 转换distributions字符串为对应Evolution对象 #

在加载YAML配置后，添加字符串到Evolution对象的映射逻辑：

# 在generate_testset函数中，加载configs后添加
evolution_map = {
    "simple": simple,
    "reasoning": reasoning,
    "multi_context": multi_context
}
# 将配置中的字符串转为对应的Evolution实例
configs["distributions"] = [evolution_map[evo] for evo in configs["distributions"]]

2. 替换tqdm导入方式 #

将脚本中的tqdm.notebook替换为普通tqdm：

# 原导入
# from tqdm.notebook import tqdm
# 替换为
from tqdm import tqdm

修复后的完整generate_testset函数 #

def generate_testset(configs_file):
    """Generate RAGAS synthetic test set"""
    with open(configs_file) as f:
        configs = yaml.safe_load(f)

    # 新增：转换distributions字符串为Evolution对象
    evolution_map = {
        "simple": simple,
        "reasoning": reasoning,
        "multi_context": multi_context
    }
    configs["distributions"] = [evolution_map[evo] for evo in configs["distributions"]]

    collection_name = configs["collection_name"]
    filegroup_name = configs["filegroup_name"]
    version = configs["version"]

    files = get_pdf_files(collection_name, filegroup_name)
    print(f"Number of files: {len(files)}")

    docs = []
    for file in tqdm(files):
        doc_ = PDFPlumberLoader(
            str(file),
            text_kwargs={
                "x_tolerance": configs["x_tolerance"],
                "y_tolerance": configs["y_tolerance"],
            },
        ).load()
        doc_ = combine_all_page(doc_)
        docs.extend(doc_)
    print(f"Number of documents: {len(docs)}")

    for doc_ in docs:
        doc_.metadata["filename"] = doc_.metadata["source"]

    splitted_testset_docs = docs

    # RAGAS to generate a synthetic testset
    generator = TestsetGenerator.with_openai()

    testset = generator.generate_with_langchain_docs(
        splitted_testset_docs,
        test_size=configs["test_size"],
        with_debugging_logs=True,
        distributions=configs["distributions"],
    )

验证 #

修改后，无论是终端运行脚本还是在Notebook中导入脚本调用，代码均可正常生成测试集，不再触发AttributeError。

内容的提问来源于stack exchange，提问作者Nicolas Ferland