Create Agent

Question

Hi.
I am making a bot with agents and tools.
My Tools are like this:
def pm_agent(idea):
return llm(pm_prompt.format_messages(idea=idea)).content
And then i created list of tools, then my memory and at last the agent that is like this:
agents = initialize_agent(
tools,
llm,
agent='zero-shot-react-description',
memory=memory,
verbose=True,
handle_parsing_errors=True
)
And then i run it with an idea but it entered a infinity loop of Think and Action until my GPU memory can not continue without any answer.
I have a question i used Qwen4B could it be cause of my model?
i saw many codes in langchain that use openai models.

Answer 1

Yes, your infinite loop is almost certainly due to the model not following the expected ReAct format. OpenAI’s models work because they’re trained for it; Qwen-4B is not. You can fix it by either enforcing stricter output parsing, giving a stronger system prompt, or switching to a model instruction-tuned for tool-using agents.
Good Luck!

Answer 2

if other model works, then maybe.
did you try any other model?
which tool did you use? : ollama, hugging face. lmstudio, .... ?