The problem in GPTQ Vicuna and GGML Vicuna
#8
by
Feng7815
- opened
GPTQ Vicuna has a problem that AI will asks and answers itself . Does GGML Vicuna have this question？（use this model by llama）
It's been a problen. My rev1 conversions had the EOS end token fix for a while. It has now been implemented officially.
That means the new v1.1 GPTQ models should be fixed too.
I have also updated all my 13B and 7B GGML models. You can download the new ones here: https://huggingface.co/eachadea/ggml-vicuna-13b-1.1