Llama.cpp fixes have been merged, requires gguf regen
#5
by
RamoreRemora
- opened
A few hours ago, llama.cpp introduced fixes for gemma2.
All GGUF quants have to be regenerated for both 9b and 27b, unfortunately:
https://github.com/ggerganov/llama.cpp/pull/8197
Thanks again for your amazing work!
That don't have to be because they'll be populated with default values - but I'm doing it anyways for completeness and because imatrix should improve in quality with these changes
they've been remade!
My download was at 97% on ~1Mbs connection 🥹 - But thanks for your work! 🫡