can we have a 30B scale of this model?
#9
by
Tonight223
- opened
This model works well with different task, can we have a 30B scale of this model? 30B version of this maybe the best 30B model ever.
We have to wait for CarperAI (or stabilityai) since this is a (really good) optimization from his 13B model. The problem is that this is really expensive and take too long, I guess the next model that we'll see it's a 15B.
Apparently there are some groups working on 30B models. For example I believe the LmSys team, who made Vicuna, are working on a 30B.
But as Vordlex says, it's time consuming and expensive in GPU resources. It will cost thousands of $ to train a 30B when renting GPUs by the hour.
TheBloke
changed discussion status to
closed