基于Transformers+Gradio+FAISS的聊天机器人出现TypeError问题求助

你遇到的TypeError: string indices must be integers, not 'tuple'错误，核心原因有3个：

• 直接将HuggingFace原生模型传入FAISS，不符合LangChain的Embeddings接口要求，导致检索返回结果格式异常
• 错误调用ConversationBufferMemory不存在的update_history方法，且返回值不符合Gradio要求
• 错误访问文档对象的content属性，LangChain的Document对象实际用page_content存储内容

以下是修正后的完整代码，修复点已标注在注释中：

import torch
from transformers import (
    DistilBertTokenizer,
    DistilBertModel,
    DistilBertForQuestionAnswering,
)
import gradio as gr
from langchain.memory import ConversationBufferMemory
from langchain_community.vectorstores import FAISS
# 引入HuggingFaceEmbeddings，用于包装原生模型
from langchain_community.embeddings import HuggingFaceEmbeddings
import pathlib
import logging

# Set logging
logging.basicConfig(level=logging.INFO)

# Initialize tokenizer and models
tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
embedding_model = DistilBertModel.from_pretrained("distilbert-base-uncased")
qa_model = DistilBertForQuestionAnswering.from_pretrained(
    "distilbert-base-uncased-distilled-squad"
)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
embedding_model.to(device)
qa_model.to(device)

# 修复点1：用HuggingFaceEmbeddings包装原生模型，符合LangChain接口要求
embeddings = HuggingFaceEmbeddings(
    model_name="distilbert-base-uncased",
    model_kwargs={"device": device},
    tokenizer_kwargs={"padding": True, "truncation": True, "max_length": 512}
)

# Load FAISS index
index_path = pathlib.Path("./saved-index-faiss")  # Update the path as necessary
embeddings_db = FAISS.load_local(
    index_path, embeddings, allow_dangerous_deserialization=True
)
retriever = embeddings_db.as_retriever(
    search_kwargs={"k": 3}
)  # Adjust 'k' as necessary for your context retrieval

# Setup memory for conversation
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)


# Define a function to answer questions using the provided context
def answer_question(question, context):
    inputs = tokenizer(
        question,
        context,
        return_tensors="pt",
        truncation=True,
        padding="max_length",
        max_length=512,
    )
    inputs = {k: v.to(device) for k, v in inputs.items()}

    with torch.no_grad():
        outputs = qa_model(**inputs)

    answer_start = torch.argmax(outputs.start_logits)
    answer_end = torch.argmax(outputs.end_logits) + 1

    answer = tokenizer.decode(
        inputs["input_ids"][0][answer_start:answer_end], skip_special_tokens=True
    )
    return answer


# Define the Gradio function
def ask(message, history):
    if not isinstance(message, str):
        message = str(message)  # Convert to string if it's not already

    print("Input message:", message)

    # Fetch context based on the retrieval model
    results = retriever.invoke(message)
    # 修复点2：访问Document对象的page_content属性，而非content
    context = results[0].page_content if results else "No relevant information found."
    print("Retrieved context:", context)

    # 修复点3：使用正确的save_context方法更新对话记忆，且不将其作为返回值
    answer = answer_question(message, context)
    memory.save_context({"input": message}, {"output": answer})

    # 返回回答字符串，符合Gradio要求
    return answer


# Create the Gradio interface
io = gr.ChatInterface(
    fn=ask,
    chatbot=gr.Chatbot(height=400),
    textbox=gr.Textbox(placeholder="Ask Away! ", container=False, scale=6),
    title="WikiBuddy",
    description="Ask me any question",
    theme="soft",
    examples=[
        "Who is David Beckham?",
        "Is Bitcoin Dead?",
        "What club does Christiano Ronaldo plays for?",
    ],
    retry_btn=None,
    undo_btn="Delete Previous",
    clear_btn="Clear",
    analytics_enabled=True,
    fill_height=True,
)
# Run the Gradio app
io.launch()

1. Embeddings适配：LangChain的FAISS需要使用LangChain定义的Embeddings接口实现，而非直接传入HuggingFace原生模型，用HuggingFaceEmbeddings包装后，确保索引加载和检索时格式正确。
2. 文档属性修正：LangChain检索返回的Document对象，内容存储在page_content属性中，而非你代码中的content，这是触发字符串索引错误的直接原因。
3. 对话记忆修正：ConversationBufferMemory没有update_history方法，正确更新记忆的方式是使用save_context；同时Gradio的ChatInterface要求函数仅返回回答字符串，不需要返回内存操作结果，避免额外格式错误。

内容的提问来源于stack exchange，提问作者Ghulam Mustafa