Model comparison

Fun-CosyVoice3-0.5B-2512 vs neutts-air

These two publish different quantization sets; the table below has the exact sizes.

From the file· summed bytes, KV per layer

Side by side

Fun-CosyVoice3-0.5B-2512neutts-air
Parameters748M
Architecturecosyvoice3-campplusqwen2
Layers24
Native context32,768
Mixture of expertsnono
Quantizations published36
Smallest quantization0.34 GiB0.43 GiB
Q4_K_M0.43 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextFun-CosyVoice3-0.5B-2512neutts-airRatio
4,0960.05 GiB
8,1920.09 GiB
16,3840.19 GiB
32,7680.38 GiB
65,5360.75 GiB
131,0721.50 GiB