Hacker News

habedi0
Show HN: Cogitator – A Python Toolkit for Chain-of-Thought Prompting github.com

Hi everyone,

I made an open-source Python toolkit/library, named Cogitator, to make it easier to try and use different chain-of-thought (CoT) reasoning methods. The project is at the beta stage, but it supports using models provided by OpenAI and Ollama. It includes implementations for Cot strategies and frameworks like Self-Consistency, Tree of Thoughts, and Graph of Thoughts.

GitHub link of the project: https://github.com/habedi/cogitator


manifoldgeo4 hours ago

Will there be a follow-up toolkit for Artificial General Intelligence called Agitator?

Dumb jokes aside, I took a look at your GitHub page, and this is exactly what I've been looking for when I do local LLM work. Cogitator seems like a nice, pythonic approach vs. using the raw `ollama run` command, esp. given the focus on chain of thought. I think I'll start using this tool. Nice work!

habedi0op4 hours ago

Thanks.

tomaytotomato6 hours ago

Very nice, will there be support for other models in the future?

This technology is not heretical, praise the Omnissiah!

habedi0op6 hours ago

Lol. No. Not heretical at all.

I might add support for other model providers (like Google and Azure) in the future. Although I'm trying to keep the scope of the project very small because it's easier for me to maintain it.

Anyway, I think adding new LLM providers is pretty straightforward if you want to do it yourself. You just need to implement the API of the BaseLLM (see the `cogitator/model/base.py` file) for your provider. After that, you just use it like how you use OllamaLLM or OpenAILLM.

nico9 hours ago

Very interesting. I saw the examples and would have loved to see the results, maybe some text showing the whole process or a little gif/video

Great work

habedi0op8 hours ago

Sounds like a great idea. In the next release, I'll add a visualization for how things work and related to each other, and possibly also include some benchmark results.

nico8 hours ago

Cool, yes benchmark results are great to show

Also, are you using this tool as part of another project? It’d be interesting to see what the main applications of CoT prompting are (the examples are great but a little basic)

habedi0op7 hours ago

I'm not using it in a larger project at the moment. The examples right now are mainly included to help people get started quickly. About the applications, they are somewhat context-dependent, but I might add one or two larger examples later if I have the time.

I guess CoT prompting could be used for ARC Prize (https://arcprize.org)

hn-front (c) 2024 voximity
source