prebuilt React agent to allow changing ChatModel parameters such as temperature in the invoke or stream calls.

[bzadon]

Unlike ChatModel.invoke, prebuilt React agent .invoke() does not allow passing any additional kwargs that flow into the underlying ChatModel in order to customize parameters such as temperature.

[hinthornw]

from langchain.chat_models import init_chat_model
from langgraph.prebuilt import create_react_agent

llm = init_chat_model("openai:gpt-4o", temperature=0.7)

create_react_agent(llm, tools=[]).invoke({"messages": "It's getting hot in here"})

[bzadon]

yes, but can i do this?

graph = create_react_agent(llm, tools[])
cold = graph.invoke({"messages":"I am freezing"}, temperature=0.0)
hot = graph.invoke({"messages":"I am burning up"}, temperature=0.9)

[vbarda]

this is not possible at the moment with create_react_agent. you could implement your own version and dynamically configure the model inside call_model node. will look into this for the prebuilt react agent

[bzadon]

tracing through the call stack, the kwargs from invoke() just need to get passed all the way through. Right now only langchain.invoke is doing that. not the react agent graph.invoke. TIA!!

[ccurme]

I think you can use configuration for this:

from langchain_core.tools import tool
from langchain.chat_models import init_chat_model
from langgraph.prebuilt import create_react_agent


llm = init_chat_model(
    "openai:gpt-4o-mini",
    configurable_fields=("temperature",),
)


@tool
def get_weather(location: str):
    """Get the weather."""
    return "It's sunny."


agent = create_react_agent(llm, [get_weather])
agent.invoke(
    {"messages": "What's the weather in Boston?"},
    {"configurable": {"temperature": 100}},  # raises BadRequestError
)

[bzadon]

agent.invoke(
    {"messages": "What's the weather in Boston?"},
    {"configurable": {"temperature": 100}}

unfortunately that didn't work for my chat model (AzureChatOpenAI):

  File ".../site-packages/langchain_openai/chat_models/base.py", line 790, in _generate
    response = self.client.create(**payload)
  File ".../site-packages/openai/_utils/_utils.py", line 279, in wrapper
    return func(*args, **kwargs)
TypeError: Completions.create() got an unexpected keyword argument 'configurable_fields'
During task with name 'agent' and id 'ec7eef4f-9ffa-f777-9592-a5c2a8788e8e'

i traced through the call stack and the temperature parameter doesn't get passed correctly as it would if calling invoke directly on the chat model.

[ccurme]

Can you try updating your langchain packages?

If that doesn't work, could you share more details about your environment? You can run

python -m langchain_core.sys_info