The model often answers the question over and over again with no overflow of video memory
#2
by
xldistance
- opened
I wonder if the question is too long
Try the new merge instead, should be better: https://huggingface.co./brucethemoose/Yi-34B-200K-DARE-merge-v5
Also, all YI models are kind of unstable and need MinP sampling but a good context to function well.