What model is Codeium based on Is a a

[stray leaf]

Autocomplete model is trained from scratch entirely in-house. Chat uses OpenAI gpt-3.5-turbo but we are working on training our own in-house chat model.

[chrome idol]

That's cool. Is the autocomplete model lightweight enough to eventually add a feature to run it locally? (optionally)

[stray leaf]

Very unlikely unfortunately

[chrome idol]

Well, good luck on the custom model training.

[stray leaf]

Thanks!

[chrome idol]

BTW, do you (or will you) have finetuning support. I would like to finetune it to work on a currently unsupported programming language.

[stray leaf]

We only support fine tuning for enterprises at the moment.

[chrome idol]

Oh ok.

[chrome idol]

How's that going? Are you just finetuning an existing LLM or creating your own from scratch? How big is it? Is it performing better than GPT-3.5-turbo yet?