Prompt must accept context as an input variable

[minnahu23]

I am trying to run the example on Quickstart page, a history aware retriever, https://python.langchain.com/docs/get_started/quickstart, but get an error: ValueError: Prompt must accept context as an input variable. Received prompt with input variables: ['chat_history', 'input']

How do I fix it? Thank you!

Here is my code:
loader = UnstructuredFileLoader("./knowledge.txt")
docs = loader.load()
print(docs)
documents = RecursiveCharacterTextSplitter(
chunk_size=1000, chunk_overlap=200
).split_documents(docs)
vector = FAISS.from_documents(documents, OpenAIEmbeddings())
retriever = vector.as_retriever()

prompt_retriever = ChatPromptTemplate.from_messages([
    MessagesPlaceholder(variable_name="chat_history"),
    ("user", "{input}"),
    ("user",
     "Given the above conversation, generate a search query to look up in order to get information relevant to the conversation")
])
retriever_chain = create_history_aware_retriever(llm, retriever, prompt_retriever)

prompt = ChatPromptTemplate.from_messages(
    [
        SystemMessage(
            content="Answer the user's questions based on the below context:\n\n{context}"
        ),  # The persistent system prompt
        MessagesPlaceholder(
            variable_name="chat_history"
        ),  # Where the memory will be stored.
        HumanMessagePromptTemplate.from_template(
            "{input}"
        ),  # Where the human input will injected
    ]
)

document_chain = create_stuff_documents_chain(llm, prompt)
retrieval_chain = create_retrieval_chain(retriever_chain, document_chain)

chat_history = [HumanMessage(content="Can I cancel AI Business Coach subscription?"), AIMessage(content="Yes!")]
retrieval_chain.invoke({
    "chat_history": chat_history,
    "input": "Tell me how"
})

[dosubot[bot]]

🤖

Hello @minnahu23! 👋 I'm Dosu, a friendly bot here to assist while we await a human maintainer. I'm equipped to help solve bugs, answer questions, and guide you to contribute to this project. Let's get your issue sorted out!

The error message you're encountering, "ValueError: Prompt must accept context as an input variable", indicates that the prompt configuration you're using doesn't include "context" as an input variable. In LangChain, the "context" variable is used to provide the history or context of the conversation to the language model.

However, looking at your code, it seems like you're not using the "context" variable in your prompt configurations. Instead, you're using "chat_history" and "input" as your input variables. This is likely the cause of the error.

To resolve this issue, you need to add "context" to the list of input variables in your prompt configuration. Here's how you can modify your ChatPromptTemplate.from_messages:

prompt = ChatPromptTemplate.from_messages(
    [
        SystemMessage(
            content="Answer the user's questions based on the below context:\n\n{context}"
        ),  # The persistent system prompt
        MessagesPlaceholder(
            variable_name="context"
        ),  # Where the memory will be stored.
        HumanMessagePromptTemplate.from_template(
            "{input}"
        ),  # Where the human input will injected
    ]
)

In this modified configuration, the "context" variable is included in the list of input variables and is also used in the template. This will allow the prompt to accept the context of the conversation as an input.

Please try this modification and let me know if it resolves your issue. If you encounter any other problems, feel free to ask.

Sources

• docs/docs/modules/model_io/prompts/simple_prompt.yaml
• libs/core/langchain_core/prompts/chat.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[minnahu23]

@dosu-bot: I updated it to variable_name="context", it solved the above error, thank you!

However I have a new error now, ValueError: variable context should be a list of base messages, how do I fix it? thank you

[dosubot[bot]]

🤖

I'm glad to hear that you were able to resolve the previous error. Now, regarding the new error you're encountering, "ValueError: variable context should be a list of base messages", it seems that the context variable you're passing to the invoke() function is not in the correct format.

In LangChain, the context variable should be a list of BaseMessage objects. Each BaseMessage object represents a message in the conversation and can be either a HumanMessage or an AIMessage. Here's an example of how you might create a context variable:

from langchain_core.messages import HumanMessage, AIMessage

context = [
    HumanMessage(content="Can I cancel AI Business Coach subscription?"),
    AIMessage(content="Yes!"),
]

In your case, it seems like you're trying to pass a string or JSON as the context. You'll need to convert this to a list of BaseMessage objects first. You can do this by creating HumanMessage or AIMessage objects for each message in your string or JSON and adding them to the list. Here's an example of how you might do this:

from langchain_core.messages import HumanMessage, AIMessage

# Assuming `json_context` is your JSON context
context = []
for message in json_context:
    if message['type'] == 'human':
        context.append(HumanMessage(content=message['content']))
    elif message['type'] == 'ai':
        context.append(AIMessage(content=message['content']))

After this, you can use context as the input to the invoke() function.

For more information about the BaseMessage object in LangChain, you can refer to the source code.

I hope this helps! If you have any other questions or run into any other issues, feel free to ask.

Sources

• ValueError: variable chat_history should be a list of base messages, got [HumanMessage(content='input message'), "output response"]
• libs/core/langchain_core/beta/runnables/context.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[minnahu23]

@dosu-bot
To my understanding, the context is passed in by the retriever_chain, not me. Appreciate the answer, but I don't think it will fix the problem. Any other suggestions? Thank you!
retrieval_chain = create_retrieval_chain(retriever_chain, document_chain)

[dosubot[bot]]

🤖

Based on the error message you're receiving, it seems like the context variable you're passing into the retrieval_chain.invoke() method is not a list of BaseMessage instances. In the LangChain framework, the context variable should be a list of BaseMessage instances.

Here's how you can create a list of BaseMessage instances:

from libs.core.langchain_core.messages.base import BaseMessage

# Create a list of BaseMessage instances
context = [
    BaseMessage(content="Hello, world!", type="Greeting"),
    BaseMessage(content="How are you?", type="Question"),
]

Please replace the content and type values with the ones that are relevant to your application.

However, if you're saying that the context is passed in by the retriever_chain, then it might be an issue with how the retriever_chain is creating the context. Without more specific details about how the retriever_chain is implemented, it's hard to provide a more precise solution.

If you're still facing issues, could you please provide more specific details about the retriever_chain and how it's creating the context? This would help in identifying the root cause of the issue and providing a more accurate solution.

Sources

• libs/core/langchain_core/messages/base.py
• libs/core/langchain_core/beta/runnables/context.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[minnahu23]

@dosu-bot thank you, here is the code for retriever_chain, thank you

loader = UnstructuredFileLoader("./knowledge.txt")
docs = loader.load()
print(docs)
documents = RecursiveCharacterTextSplitter(
chunk_size=1000, chunk_overlap=200
).split_documents(docs)
vector = FAISS.from_documents(documents, OpenAIEmbeddings())
retriever = vector.as_retriever()

prompt_retriever = ChatPromptTemplate.from_messages([
MessagesPlaceholder(variable_name="chat_history"),
("user", "{input}"),
("user",
"Given the above conversation, generate a search query to look up in order to get information relevant to the conversation")
])
retriever_chain = create_history_aware_retriever(llm, retriever, prompt_retriever)

[dosubot[bot]]

🤖

Based on the code you've provided, it seems like the error might be due to the format of the chat history. In LangChain, the chat history should be a list of CHAT_TURN_TYPE which is defined as a Union of Tuple[str, str] and BaseMessage.

Here is an example of a valid chat history:

chat_history = [
    BaseMessage(type="human", content="Hello, how are you?"),
    BaseMessage(type="ai", content="I'm good, thank you!"),
    ("What's the weather like?", "It's sunny today.")
]

Please ensure that the chat history you are providing is a list where each element is either an instance of BaseMessage or a tuple of two strings.

Regarding the create_history_aware_retriever function, it expects three input parameters:

1. llm: An instance of a class that behaves like a language model. The exact type is LanguageModelLike.
2. retriever: An instance of a class that behaves like a retriever. The exact type is RetrieverLike. This object should take a string as input and output a list of Documents.
3. prompt: An instance of BasePromptTemplate. This is used to generate the search query for the retriever.

The function returns an object of type RetrieverOutputLike.

If you are still encountering the error after ensuring the chat history is in the correct format, there may be an issue with the BaseMessage class or the _get_chat_history function. In this case, you may need to debug these components to find the source of the error.

For more details, you can refer to the source code of LangChain:

• libs/langchain/langchain/chains/conversational_retrieval/base.py
• libs/langchain/langchain/chains/history_aware_retriever.py

I hope this helps! If you have any more questions or if something is not clear, please let me know.

Sources

• libs/langchain/langchain/chains/conversational_retrieval/base.py
• libs/langchain/langchain/chains/history_aware_retriever.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[minnahu23]

@dosu-bot: it has nothing to do with chat_history.

how does retriever_chain pass context to document_chain in the sample code?

retriever_chain = create_history_aware_retriever(llm, retriever, prompt_retriever)
document_chain = create_stuff_documents_chain(llm, prompt)
retrieval_chain = create_retrieval_chain(retriever_chain, document_chain)

[dosubot[bot]]

🤖

Based on the code you've provided, it seems like the retriever_chain is created using the create_history_aware_retriever function. This function creates a chain that takes conversation history and returns documents. If there is no chat_history, then the input is just passed directly to the retriever. If there is chat_history, then the prompt and Language Model (LLM) will be used to generate a search query. That search query is then passed to the retriever.

The retriever_chain passes the context to the document_chain through the RunnablePassthrough.assign method. In the provided code, retrieval_docs is a Runnable object that retrieves documents. It is assigned to the context key of the RunnablePassthrough object. This means that the output of retrieval_docs (the retrieved documents) will be passed as the context to the next step in the chain.

Then, combine_docs_chain is assigned to the answer key of the RunnablePassthrough object. This means that combine_docs_chain will receive the context (the retrieved documents) as its input.

Here is the relevant code:

retrieval_chain = (
    RunnablePassthrough.assign(
        context=retrieval_docs.with_config(run_name="retrieve_documents"),
    ).assign(answer=combine_docs_chain)
).with_config(run_name="retrieval_chain")

In this code, RunnablePassthrough.assign is used to assign the output of retrieval_docs to the context key, and the combine_docs_chain to the answer key. This way, the retrieved documents are passed as the context to the combine_docs_chain.

You can find more details in the libs/langchain/langchain/chains/retrieval.py and libs/langchain/langchain/chains/history_aware_retriever.py files in the LangChain repository.

I hope this helps! If you have any more questions, feel free to ask.

Sources

• libs/langchain/langchain/chains/retrieval.py
• libs/langchain/langchain/chains/history_aware_retriever.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[TomikusMaximus]

Hi: I'm offering a little prompt to help you fix future bugs. I believe it will be useful.

---

You are an AI assistant tasked with analyzing a code snippet and answering a question about it. Here's the code snippet you need to analyze:

<code_snippet>
{{CODE_SNIPPET}}
</code_snippet>

Your task is to carefully analyze this code and answer the following question:

{{QUESTION}}

To analyze the code:

1. Read through the entire code snippet carefully.
2. Identify the main components and their functions.
3. Pay attention to the variable names, function calls, and overall structure.
4. Note any important details or patterns that might be relevant to the question.

To answer the question:

1. Consider how the code snippet relates to the question asked.
2. Use your understanding of the code to formulate a clear and concise answer.
3. If necessary, refer to specific parts of the code to support your answer.
4. Ensure your explanation is accurate and based solely on the information provided in the code snippet.

Please provide your answer in the following format:

[Your detailed analysis of the code snippet]

[Your clear and concise answer to the question]

Remember to base your analysis and answer only on the information provided in the code snippet. Do not make assumptions about any external libraries, functions, or context that isn't explicitly shown in the code.