点击Streamlit按钮触发Chroma ValueError：Expected IDs非空列表排查

错误详情 #

**ValueError:** Expected IDs to be a non-empty list, got []

**Traceback:**
File "C:\Users\scite\Desktop\HAMBOTAI\HAMBotAI\HAMBotAI\homehambotai.py", line 96, in app
    db = Chroma.from_documents(texts, embeddings)
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\langchain_community\vectorstores\chroma.py", line 771, in from_documents
    return cls.from_texts(
           ^^^^^^^^^^^^^^^
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\langchain_community\vectorstores\chroma.py", line 729, in from_texts
    chroma_collection.add_texts(
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\langchain_community\vectorstores\chroma.py", line 324, in add_texts
    self._collection.upsert(
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\chromadb\api\models\Collection.py", line 449, in upsert
    ) = self._validate_embedding_set(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\chromadb\api\models\Collection.py", line 512, in _validate_embedding_set
    valid_ids = validate_ids(maybe_cast_one_to_many_ids(ids))
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\scite\AppData\Roaming\Python\Python311\site-packages\chromadb\api\types.py", line 228, in validate_ids
    raise ValueError(f"Expected IDs to be a non-empty list, got {ids}")

相关代码片段 #

if 'processed' in query_params:
    # Create a temporary text file
    with tempfile.NamedTemporaryFile(mode="w", delete=False, suffix=".txt") as temp_file:
        temp_file.write(text)
        temp_file_path = temp_file.name

    # load document
    loader = TextLoader(temp_file_path)
    documents = loader.load()
    # split the documents into chunks
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    texts = text_splitter.split_documents(documents)
    # select which embeddings we want to use
    embeddings = OpenAIEmbeddings()
    # ids =[str(i) for i in range(1, len(texts) + 1)]
    # create the vectorestore to use as the index
    db = Chroma.from_documents(texts, embeddings)
    # expose this index in a retriever interface
    retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 2})
    # create a chain to answer questions
    qa = ConversationalRetrievalChain.from_llm(OpenAI(), retriever)
    chat_history = []
    # query = "What's the Name of patient and doctor as mentioned in the data?"
    # result = qa({"question": query, "chat_history": chat_history})
    # st.write("Patient and Doctor name:", result['answer'])
    #
    # chat_history = [(query, result["answer"])]
    query = "Provide summary of medical and health related info from this data in points, every point should be in new line (Formatted in HTML)?"
    result = qa({"question": query, "chat_history": chat_history})
    toshow = result['answer']
    # chat_history = [(query, result["answer"])]
    # chat_history.append((query, result["answer"]))
    # print(chat_history)

    st.title("Data Fetched From Your Health & Medical Reports")
    components.html(
        f"""
        {toshow}
        """,
        height=250,
        scrolling=True,
    )

    if st.button('Continue to Questionarrie'):
        st.write('Loading')
    st.text("(OR)")
    if st.button('Chat with BotAI'):
        st.title("Chat with BotAI")

核心原因 #

Streamlit的按钮点击会触发整个脚本重新运行：

• 点击按钮时，URL的query_params仍保留processed标识，代码会再次进入if 'processed' in query_params分支
• 此时text变量已丢失有效内容（没有重新生成/上传文本），导致加载的文档为空，拆分后的texts是空列表
• Chroma调用from_documents时，空的texts会生成空的ID列表，触发校验错误

解决方案 #

1. 用Streamlit会话状态缓存处理结果 #

把向量库、QA链、摘要结果等存入st.session_state，避免重复执行文档处理逻辑：

if 'processed' in query_params:
    # Create a temporary text file
    with tempfile.NamedTemporaryFile(mode="w", delete=False, suffix=".txt") as temp_file:
        temp_file.write(text)
        temp_file_path = temp_file.name

    # load document
    loader = TextLoader(temp_file_path)
    documents = loader.load()
    # split the documents into chunks
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    texts = text_splitter.split_documents(documents)
    
    # 新增：校验texts是否为空
    if not texts:
        st.error("上传的文本内容为空，无法生成向量库")
    else:
        embeddings = OpenAIEmbeddings()
        db = Chroma.from_documents(texts, embeddings)
        retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 2})
        qa = ConversationalRetrievalChain.from_llm(OpenAI(), retriever)
        chat_history = []
        query = "Provide summary of medical and health related info from this data in points, every point should be in new line (Formatted in HTML)?"
        result = qa({"question": query, "chat_history": chat_history})
        toshow = result['answer']
        
        # 缓存结果到会话状态
        st.session_state['toshow'] = toshow
        st.session_state['qa_chain'] = qa

# 从会话状态读取结果，避免重复处理
if 'toshow' in st.session_state:
    st.title("Data Fetched From Your Health & Medical Reports")
    components.html(
        f"""
        {st.session_state['toshow']}
        """,
        height=250,
        scrolling=True,
    )

    if st.button('Continue to Questionarrie'):
        st.write('Loading')
        # 可在这里切换会话状态，比如st.session_state['page'] = 'questionnaire'
    st.text("(OR)")
    if st.button('Chat with BotAI'):
        st.title("Chat with BotAI")
        # 可使用缓存的qa_chain进行对话

2. 按钮点击后更新页面状态 #

点击按钮后，通过会话状态标记当前页面，避免重新进入processed分支：

if st.button('Continue to Questionarrie'):
    st.session_state['current_page'] = 'questionnaire'
    st.experimental_rerun()

# 在脚本开头判断页面状态
if 'current_page' in st.session_state and st.session_state['current_page'] == 'questionnaire':
    # 渲染问卷页面逻辑
    st.write("问卷页面")
else:
    # 原有的文档处理和结果展示逻辑
    pass

内容的提问来源于stack exchange，提问作者SCITECHE