It would be valuable if we tried to train on a fully open-source base model, like OLMo 2. This would make the resulting model fully open source and reproducible from the ground up.
· Sign up or log in to comment