gguf ?
gguf version?
Hi
@celsowm
! Since this model is hot-off-the-press, we don't have inference support in llama.cpp yet. I'm actively working on it, but since this is one of the first major models using a hybrid-recurrent architecture, there are a number of in-flight architectural changes in the codebase that need to all meet up to get this supported. We'll keep you posted!
Hi @celsowm ! Since this model is hot-off-the-press, we don't have inference support in
llama.cppyet. I'm actively working on it, but since this is one of the first major models using a hybrid-recurrent architecture, there are a number of in-flight architectural changes in the codebase that need to all meet up to get this supported. We'll keep you posted!
thanks for your explanation !
I hope that time granite can increase its score on this benchmark: https://huggingface.co/datasets/celsowm/legalbench.br
We definitely expect the model quality to improve beyond this preview. So far, this preview checkpoint has been trained on ~2.5T tokens, but it will continue to train up to ~15T tokens before final release.