Model comparison

OmniVoice vs Fun-CosyVoice3-0.5B-2512

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

From the file· summed bytes, KV per layer

Side by side

OmniVoiceFun-CosyVoice3-0.5B-2512
Parameters613M
Architectureomnivoice-lmcosyvoice3-campplus
Layers
Native context
Mixture of expertsnono
Quantizations published73
Smallest quantization0.56 GiB0.34 GiB
Q4_K_M0.61 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextOmniVoiceFun-CosyVoice3-0.5B-2512Ratio
4,096
8,192
16,384
32,768
65,536
131,072