/generate throws KeyError: 'description'

[labarba]

Description

Trying /generate for the first time in our JupyterHub, so I don't know if this is a bug or an installation issue on our side. The command gives an error: "Sorry, something went wrong and I wasn't able to index that path." with:

Traceback (most recent call last):
  File "/opt/conda/lib/python3.11/site-packages/jupyter_ai/chat_handlers/base.py", line 39, in process_message
    await self._process_message(message)
  File "/opt/conda/lib/python3.11/site-packages/jupyter_ai/chat_handlers/generate.py", line 241, in _process_message
    await fill_outline(outline, llm=self.llm, verbose=True)
  File "/opt/conda/lib/python3.11/site-packages/jupyter_ai/chat_handlers/generate.py", line 190, in fill_outline
    generate_code(section, outline["description"], llm=llm, verbose=verbose)
                           ~~~~~~~^^^^^^^^^^^^^^^
KeyError: 'description'

Reproduce

Any call to \generate in the chat interface produces the same error.

Expected behavior

Expected a new generated notebook to be created.

Context

jupyter-ai                2.3.0                    pypi_0    pypi
jupyter-ai-magics         2.3.0                    pypi_0    pypi
jupyterlab                4.0.6              pyhd8ed1ab_0    conda-forge

[welcome]

Thank you for opening your first issue in this project! Engagement like this is essential for open source projects! 🤗

If you haven't done so already, check out Jupyter's Code of Conduct. Also, please try to follow the issue template as it helps other other community members to contribute more effectively.

You can meet the other Jovyans by joining our Discourse forum. There is also an intro thread there where you can stop by and say Hi! 👋

Welcome to the Jupyter community! 🎉

[labarba]

I've been staring at code for hours, but could not find a clue to this mystery—the function generate_outline() receives a description as input, but when called by _generate_notebook() it gets the input variable prompt instead, which it looks like is created by PromptTemplate() from langchain.prompts. That has input_variables=["description"], but I cannot find where this may come from.

[srdas]

@labarba Are you using just /generate or the slash command followed by a prompt for what you want to generate. I tried using /generate <prompt> and it works well, as shown below.

I have also tried simply using /generate and it generates a random notebook but does not throw an error. And if the slash is a forward slash, it replies that it cannot process it. Maybe a reinstall will fix the problem? Hope this helps.

[labarba]

My exact input was:

/generate a lesson to teach beginners the essence of Pandas groupby(), including how to loop over grouped dataframes.

In the next input, I tried without the /generate command, as follows:

Give me an outline for a lesson to teach beginners the essence of Pandas groupby(), including how to loop over grouped dataframes.

… and it gave me a response as expected (indicating my connection to the LLM is good via API).

This is my uni JupyterHub installation, so not under my direct control. The admins are not keen on rebuilding the image, so I have to give them a good reason.

[srdas]

@labarba It's good that your LLM API seems to be working just fine.
I tried your exact inputs (see below) and I suspect using /generate does a better job, if you look at the results with and without the slash command (but that's a judgment call). The left hand side panel below does not use /generate and the right hand side is the notebook generated by the /generate command.

It sounds like a fresh install will solve this issue for you. Hope this helps.

[srdas]

@labarba I went ahead and reinstalled everything (including jupyterhub) in a new environment on a linux machine and tested if jupyter-ai works as expected after connecting remotely to jupyterhub on that machine. I can confirm it all works well. (I am using JLab v4.1.6.) and the latest version of jupyter-ai.

[labarba]

ha ha, meanwhile, I went ahead and installed jupyter-ai and dependencies on my local machine and it worked well with the /generate command, so while I convince the admin of our JupyterHub to reinstall I can do what I wanted to do on localhost!

[srdas]

That's good to hear, and confirms that it is easier to solve a bug than bureaucracy! I'll go ahead and close the issue.