Supported models

[veeral-patel]

Thanks for this project! Which models are supported? Is it just GPT4 or others too?

[paolorechia]

Hi, @veeral-patel, this was built with local LLMs, using either text-generation-webui (https://github.com/oobabooga/text-generation-webui) or a HTTP server I wrote a while ago, that serves models from Hugging Face (https://github.com/paolorechia/learn-langchain).

That being said, it's not complex to write a new client.

For instance, see the client for the text-generation-webui:
https://github.com/paolorechia/code-it/blob/40b9ed946019bb24fa0e172edb679df73abb62ba/code_it/models.py#LL89C1-L116C6

If you want to use with OpenAI, you'd have to write a similar client here in this module.

[veeral-patel]

Thanks! How's the performance with these models for code writing? Just curious, I've only used OpenAI

[paolorechia]

For simple examples, WizardLM performs quite ok: https://huggingface.co/TheBloke/wizardLM-7B-HF

Of course, it’s not as good as OpenAI gpt3.5-turbo.

This guy here has a nice table comparing them:
https://github.com/the-crypt-keeper/can-ai-code/tree/main

[veeral-patel]

Cool! If it's not working I can write a plug-in for GPT3.5

Also, how well does this project integrate code into a repository?

Can it modify the relevant files itself for a feature or do we need to do it by hand

[paolorechia]

Not yet - that's one of my long term goals too, but I still haven't figured out how I would tackle this feature.

[veeral-patel]

Take a look at AutoPR