will there be new merges on gemma 2?
#3
by
Slaaaaaau
- opened
Did you try using gemma 2 9b or 27b as a basis?
so far I’m trying 27b IQ4_XS 16k, it fits exactly into a 24GB vram, so far it looks like an extremely constructive model in a large context at first try (long RP with long introductory instructions), but without finetune\merges for RP.
I will once the RP and uncensored fine-tunes start rolling in, yes.
There are already some by this guy: https://huggingface.co/TheDrummer