Return_intermediate_steps Error

API
1

Hello, my today problem is:
I want to use chainlit to create nice looking chatbot, but i have two problems:

  • I want to use agents because of my data imported inside tools:
    agent_executor = AgentExecutor(agent=agent, tools=tools, memory=memory, verbose=False,
                                    return_intermediate_steps=True, max_context_length=14000)

But my context lenght is to long when i asks questions because of returning every step.
I thought its because return_intermediate_steps=True but I cant delete it because it generates error and I cant make it on False because it generates error to. What I can do to don’t show this steps in answer ?

2

I cannot find a reference to these parameters with the limited information about the software stack you are using.

If this parameter for the software is translated into the max_tokens parameter for the OpenAI call, it is set far too high. That is the space reservation for a response. That could only work on the -16k AI model, and would only allow a 2000 token input. Try a significantly lower value.

Or the parameter is meant to match the context length of the model you are using, like gpt-3.5-turbo with 4191.

3 Likes