Model comparison

csm-1b vs OmniVoice

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

From the file· summed bytes, KV per layer

Side by side

csm-1bOmniVoice
Parameters1.6B613M
Architecturecsmomnivoice-lm
Layers16
Native context2,048
Mixture of expertsnono
Quantizations published37
Smallest quantization1.34 GiB0.56 GiB
Q4_K_M0.61 GiB
Licenceccapache-2.0

KV cache by context

the term that decides long-context viability
Contextcsm-1bOmniVoiceRatio
4,0960.13 GiB
8,1920.25 GiB
16,3840.50 GiB
32,7681.00 GiB
65,5362.00 GiB
131,0724.00 GiB