Accessing chat history as input to TransformChain

[snassimr]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain_core.messages import HumanMessage, AIMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
  print(data)
  return {'output': 'aaa'}

subchain1 = TransformChain(
    input_variables=["history"], output_variables=["output"], transform=f1
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("user"),
    ]
)


chain =  prompt | subchain1

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

chain_with_history = RunnableWithMessageHistory(chain,
                                                get_session_history,
                                                input_messages_key='user',
                                                history_messages_key='history',
                                                output_messages_key="output",)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

### Description

I have some general question regarding Langchain functionality.

I'd like to use in 'history' inside subchain2 , but I can't . 
I ma getting "ValueError: Missing some input keys: {'history'}"



### System Info

System Information
------------------
> OS:  Linux
> OS Version:  #1 SMP PREEMPT_DYNAMIC Thu Jun 27 21:05:47 UTC 2024
> Python Version:  3.10.12 (main, Jul 29 2024, 16:56:48) [GCC 11.4.0]

Package Information
-------------------
> langchain_core: 0.2.39
> langchain: 0.2.16
> langchain_community: 0.2.16
> langsmith: 0.1.118
> langchain_openai: 0.1.21
> langchain_text_splitters: 0.2.4

Optional packages not installed
-------------------------------
> langgraph
> langserve

Other Dependencies
------------------
> aiohttp: 3.10.5
> async-timeout: 4.0.3
> dataclasses-json: 0.6.7
> httpx: 0.27.2
> jsonpatch: 1.33
> numpy: 1.26.4
> openai: 1.44.1
> orjson: 3.10.7
> packaging: 24.1
> pydantic: 2.9.1
> PyYAML: 6.0.2
> requests: 2.32.3
> SQLAlchemy: 2.0.34
> tenacity: 8.5.0
> tiktoken: 0.7.0
> typing-extensions: 4.12.2

[dosubot[bot]]

I found a similar discussion that addresses the issue of using 'history' inside a subchain in LangChain: How to use RunnableWithMessageHistory? ^[1].

To use 'history' inside subchain2 without getting the "ValueError: Missing some input keys: {'history'}", you need to ensure that the history_messages_key is correctly specified and that the configuration passed to the invoke method includes the necessary keys.

Here is an example of how to set up and use RunnableWithMessageHistory correctly:

from langchain_core.messages import HumanMessage, AIMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain1 = TransformChain(
    input_variables=["history"], output_variables=["output"], transform=f1
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("user"),
    ]
)

chain = prompt | subchain1

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
    input_messages_key='user',
    history_messages_key='history',
    output_messages_key="output",
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

print(response)

In this example:

• input_messages_key is set to "user".
• history_messages_key is set to "history".
• The configuration passed to invoke includes the session_id key.

This setup ensures that the history key is correctly handled, and the ValueError is avoided ^[2][3].

To continue talking to Dosu, mention @dosu.

[snassimr]

It works . Now I realize that my case is more complicated . I want to use in 'history' in later step .
Here the updated chain :

from langchain_core.messages import HumanMessage, AIMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
BaseChatMessageHistory,
InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
print(data)
return {'output': 'aaa'}

subchain1 = TransformChain(
input_variables=[], output_variables=["output"], transform=f1
)

def f2(data: dict):
print(data)
return {'output': 'aaa'}

subchain2 = TransformChain(
input_variables=["history"], output_variables=["output"], transform=f1
)

prompt = ChatPromptTemplate.from_messages(
[
MessagesPlaceholder("history"),
MessagesPlaceholder("user"),
]
)

chain = prompt | subchain1 | subchain2

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
if session_id not in store:
store[session_id] = InMemoryChatMessageHistory()
return store[session_id]

chain_with_history = RunnableWithMessageHistory(
chain,
get_session_history,
input_messages_key='user',
history_messages_key='history',
output_messages_key="output",
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
{'user': user_message},
config=config,
)

print(response)

And now I am getting ValueError: Missing some input keys: {'history'} for subchain2 .

[dosubot[bot]]

To use history in a later step of a LangChain chain without getting the "ValueError: Missing some input keys: {'history'}" for subchain2, you need to ensure that the history key is properly passed and managed throughout the chain. You can use RunnableWithMessageHistory to manage the chat message history for the chain. Here is an example of how to set it up:

from langchain_core.messages import HumanMessage, AIMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain1 = TransformChain(
    input_variables=[], output_variables=["output"], transform=f1
)

def f2(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain2 = TransformChain(
    input_variables=["history"], output_variables=["output"], transform=f2
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("history"),
        MessagesPlaceholder("user"),
    ]
)

chain = prompt | subchain1 | subchain2

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
    input_messages_key='user',
    history_messages_key='history',
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

print(response)

In this setup:

• RunnableWithMessageHistory is used to wrap the chain and manage the chat message history.
• The get_session_history function is used to retrieve or create the chat message history based on the session ID.
• The input_messages_key and history_messages_key parameters are set to ensure that the history is correctly passed and managed.

This setup ensures that the history key is properly included in the input for subchain2, preventing the "ValueError: Missing some input keys: {'history'}" error ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[snassimr]

I am still getting ValueError: Missing some input keys: {'history'} for subchain2

Should I pass history through subchain 1's output to see it in subchain 2's input ? Is this the only way ?

[dosubot[bot]]

Yes, you should pass the history through subchain 1's output to see it in subchain 2's input to avoid the "ValueError: Missing some input keys: {'history'}" error. This ensures that the necessary context is available for subchain 2.

Here's how you can modify your setup to pass the history correctly:

1. Ensure that subchain 1 outputs the history.
2. Update subchain 2 to accept the history as input.

Here's an updated version of your code:

from langchain_core.messages import HumanMessage, AIMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
    print(data)
    # Pass the history to the output
    return {'output': 'aaa', 'history': data.get('history', [])}

subchain1 = TransformChain(
    input_variables=["history"], output_variables=["output", "history"], transform=f1
)

def f2(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain2 = TransformChain(
    input_variables=["history"], output_variables=["output"], transform=f2
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("history"),
        MessagesPlaceholder("user"),
    ]
)

chain = prompt | subchain1 | subchain2

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
    input_messages_key='user',
    history_messages_key='history',
    output_messages_key="output",
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

print(response)

In this setup:

• subchain1 now outputs both output and history.
• subchain2 is configured to accept history as an input.

This ensures that the history key is correctly passed from subchain 1 to subchain 2, avoiding the "ValueError: Missing some input keys: {'history'}" error ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[snassimr]

I don't have a possibility to pass history through subchain 1 . Any other way to pass "history" from prompt to subchain2 ? Might be RunnableBranch can be useful here ?

[dosubot[bot]]

To pass "history" from the prompt to subchain2 without passing it through subchain1, you can use RunnableBranch to conditionally route the input. Here is an example of how you can achieve this:

from langchain_core.runnables import RunnableBranch, RunnableLambda, RunnablePassthrough
from langchain_core.messages import HumanMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain1 = TransformChain(
    input_variables=[], output_variables=["output"], transform=f1
)

def f2(data: dict):
    print(data)
    return {'output': 'aaa'}

subchain2 = TransformChain(
    input_variables=["history"], output_variables=["output"], transform=f2
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("history"),
        MessagesPlaceholder("user"),
    ]
)

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

# Conditionally route the input based on the presence of "chat_history"
retrieve_documents = RunnableBranch(
    (
        # Condition to check if there is no chat history
        RunnableLambda(lambda x: not x.get("chat_history", False)),
        # If no chat history, then we just pass input to subchain1
        (lambda x: x["input"]) | subchain1,
    ),
    # If chat history, then we pass inputs to subchain2
    (lambda x: x["chat_history"]) | subchain2
).with_config(run_name="chat_retriever_chain")

chain_with_history = RunnableWithMessageHistory(
    retrieve_documents,
    get_session_history,
    input_messages_key='user',
    history_messages_key='history',
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

print(response)

In this setup, RunnableBranch is used to conditionally route the input based on the presence of "chat_history". If "chat_history" is present, it is passed directly to subchain2 without going through subchain1. This approach leverages the flexibility of RunnableBranch to handle different input scenarios effectively ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[snassimr]

My case is more complicated . I need an output from subchain1 inside subchain2 along with 'history'.
So I can't split chains completely. Any other way to make 'history' accessible over the whole chain ?

[dosubot[bot]]

To make 'history' accessible over the whole chain while also passing the output from subchain1 to subchain2, you can use the RunnableBranch to handle the conditional logic and ensure that the history is passed along with the input. Here is an example based on your requirements:

from langchain_core.messages import HumanMessage
from langchain.chains import TransformChain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.chat_history import (
    BaseChatMessageHistory,
    InMemoryChatMessageHistory,
)
from langchain_core.runnables import RunnableBranch, RunnableLambda
from langchain_core.runnables.history import RunnableWithMessageHistory

def f1(data: dict):
    print(data)
    return {'output1': 'aaa'}

subchain1 = TransformChain(
    input_variables=["history"], output_variables=["output1"], transform=f1
)

def f2(data: dict):
    print(data)
    return {'output2': 'bbb'}

subchain2 = TransformChain(
    input_variables=["history", "output1"], output_variables=["output2"], transform=f2
)

prompt = ChatPromptTemplate.from_messages(
    [
        MessagesPlaceholder("history"),
        MessagesPlaceholder("user"),
    ]
)

store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
    if session_id not in store:
        store[session_id] = InMemoryChatMessageHistory()
    return store[session_id]

# Create a chain that ensures history is passed along with the input
chain = prompt | subchain1 | (lambda x: {**x, "history": x.get("history", []), "output1": x["output1"]}) | subchain2

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
    input_messages_key='user',
    history_messages_key='history',
)

user_message = [HumanMessage(content="How are you ?")]

config = {"configurable": {"session_id": "abc5"}}

response = chain_with_history.invoke(
    {'user': user_message},
    config=config,
)

print(response)

In this setup:

• subchain1 processes the input and outputs output1.
• The lambda function ensures that history and output1 are passed to subchain2.
• subchain2 then processes the input using both history and output1.

This way, the history is accessible throughout the chain, and the output from subchain1 is passed to subchain2 ^[1][2].

To continue talking to Dosu, mention @dosu.