is this using ternary packing as in kitten-tts-2-cpp?
#3
by ronjonro - opened
we can do better than q4 too, by exploiting the ternary weights. have you tried this? https://github.com/KittenML/kitten-tts-2-cpp ?
I believe audio.cpp already uses ternary weights. Anyway, I stopped maintaining this so your questions would be better on the audio.cpp github.
dignome changed discussion status to closed