langgraph/tutorials/lats/lats/

[giscus[bot]]

langgraph/tutorials/lats/lats/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/tutorials/lats/lats/

[damienbenveniste]

Pretty sure
average_reward = self.value / self.visits
is redundant!

[108krohan]

Very cool implementation! Do we also expect this implementation to be vended as another langchain_experimental agent? There are some pretty cool implementations there, for example tree of thought (which this one seems to be better than), along with reinforcement chain, plan and execute, and fallacy removal.

[jerry0li]

Totally the same code except I used a hosted ollama model to replace gpt-4o. Then the following error occurred.

   28 @as_runnable
    29 def reflection_chain(inputs) -> Reflection:
    30     tool_choices = reflection_llm_chain.invoke(inputs)
---> 31     reflection = tool_choices[0]
    32     if not isinstance(inputs["candidate"][-1], AIMessage):
    33         reflection.found_solution = False

IndexError: list index out of range

[lvlc233]

I don't know why, but I'm pretty much the same code as in the tutorial, except for llm. But there was this error:

ValidationError: 2 validation errors for AIMessage
tool_calls.0.args
Input should be a valid dictionary [type=dict_type, input_value='{"query": "house sparrow...est recorded instance"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.9/v/dict_type
tool_calls.1.args
Input should be a valid dictionary [type=dict_type, input_value='{"query": "dark-eyed jun...est recorded instance"}', input_type=str]
For further information visit https://errors.pydantic.dev/2.9/v/dict_type

The model I used was:

llm = ChatOpenAI(
model_name="Qwen/Qwen2.5-32B-Instruct",
openai_api_base="https://api.siliconflow.cn/v1",
openai_api_key=API_KEY,
)

This llm can proceed tool_call
In fact, I've seen other tutorials, but there's always this error.