Kai
As far as I understand, it is slow because for every run, a new Julia instance is started, which takes time to compile. Is there a way to just use...
I am aware of that option, but it would be nice if something like that would be implemented under the run option.
Just saw this, which resolves the second part: https://github.com/google/gemma.cpp/pull/33
Hi Eve, Nice that you are reaching out, it's always great to hear from the people who are using your code :) Also, thanks for pointing out the error in...
Thanks for pointing that out, it should be fixed now.
Do you put this into every block or just the last block on each level? And on which levels do you use cross-attention? How is it for your NAFNet modification?...