Spaces:

Duplicated from  hysts/mistral-7b

Rijgersberg

/

GEITje-7B-chat

like 14

Running on Zero

I've created a new large open Dutch language model called GEITje 7B. It's a personal project, and it was made by further pretraining Mistral 7B on 10 billion tokens of Dutch text to make it master Dutch and Dutch subjects. I hope it serves as a basis for more progress in open source Dutch LLM-land, because pickings are slim right now due to a lack of resources.

As a demonstration, I've also trained a chat-variant called GEITje-7B-chat, which is demoed in this space. Being able to try it out easily should increase the number of open source spin-off projects.

I don't have the resources to keep the GPU in this space running for long, so I was hoping the space could get some kind of community grant.