构建RAG应用遇LangChainInterface验证错误：如何配置Watsonx凭证？

错误详情 #

运行代码时触发Pydantic验证错误，具体信息如下：

pydantic.error_wrappers.ValidationError: 3 validation errors for LangChainInterface
credentials
  instance of Credentials expected (type=type_error.arbitrary_type; expected_arbitrary_type=Credentials)
model_id
  extra fields not permitted (type=value_error.extra)
project_id
  extra fields not permitted (type=value_error.extra)
Traceback:
File "D:\projects\UniGpt\APP\app.py", line 22, in <module>
    llm = LangChainInterface(
          ^^^^^^^^^^^^^^^^^^^
File "C:\Users\CHAMIKA\unigpt-env\Lib\site-packages\langchain_core\load\serializable.py", line 113, in __init__
    super().__init__(*args, **kwargs)
File "pydantic\main.py", line 341, in pydantic.main.BaseModel.__init__

问题代码 #

#import logchain dependancies 
from langchain.document_loaders import PyPDFLoader
from langchain.indexes import VectorstoreIndexCreator
from langchain.chains import RetrievalQA
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter

#bring in steamlit for UI dev
import streamlit as st

#bring in watsonx interface
from wxai_langchain.llm import LangChainInterface,  Credentials

from config import API_KEY,URL,PROJECT_ID
#setup cerdentials directly in the code
creds={
    'api_key': API_KEY,
    'url': URL,
}

#create llm using langchain 
llm = LangChainInterface(
    credentials=creds,
    model_id="granite-3-3-8b-instruct",
    params={
        'decoding_method':'sample',
        'max_new_tokens':200,
        'temprature':0.5
    },
    project_id=PROJECT_ID
)

#function to load and index the pdf document
@st.cache_resource
def load_pdf():
    pdf_name = "Resources/HandBook.pdf"
    #load the pdf
    loader = PyPDFLoader(pdf_name)
    #split the pdf into chunks
    index=VectorstoreIndexCreator(
        embedding=HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2"),
        text_splitter=RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    ).from_loaders([loader])
    return index

#setup the retrieval qa chain
chain=RetrievalQA.from_chain_type(
    llm=llm, 
    chain_type="stuff",
    retriever=load_pdf().vectorstore.as_retriever(),
    input_key="question"
)


#setup the app title
st.title("Ask UniExpert")

#setup session state message variable to hold all the old message
if "messages" not in st.session_state:
    st.session_state.messages = []

#display all the historical messages
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

#build a prompt input template to display the prompts
prompt = st.chat_input("Enter your question:")

#if the user hits enter
if prompt:
    #display the prompt
    st.chat_message("user").markdown(prompt)
    #store the user prompt in state
    st.session_state.messages.append({"role": "user", "content": prompt})
    #response from llm
    response=chain.run(prompt)
    #show the response
    st.chat_message("assistant").markdown(response)
    #store the response in state
    st.session_state.messages.append(
        {"role": "assistant", "content": response})

已尝试方案 #

• 以字典形式传递凭证
• 查找Credentials类的正确用法

解决方案 #

1. 正确实例化Credentials类 #

不能直接传字典给credentials参数，必须创建Credentials类的实例：

creds = Credentials(api_key=API_KEY, url=URL)

2. 调整model_id和project_id的传递位置 #

当前版本的wxai_langchain中，model_id和project_id不属于LangChainInterface的直接参数，需要放到params字典里。

3. 修正参数拼写错误 #

注意参数temperature的正确拼写（原代码写成了temprature）。

修改后的LLM创建代码 #

# 实例化Credentials
creds = Credentials(api_key=API_KEY, url=URL)

# 创建llm实例
llm = LangChainInterface(
    credentials=creds,
    params={
        'model_id': "granite-3-3-8b-instruct",
        'project_id': PROJECT_ID,
        'decoding_method':'sample',
        'max_new_tokens':200,
        'temperature':0.5
    }
)

把这段代码替换原代码中对应的部分，就能解决Pydantic验证错误。

内容的提问来源于stack exchange，提问作者Chamika Udayanga