Model comparison

ARK-ASR-3B vs Voxtral-Mini-4B-Realtime-2602

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

From the file· summed bytes, KV per layer

Side by side

ARK-ASR-3BVoxtral-Mini-4B-Realtime-2602
Parameters4.1B4.4B
Architecturearkasrvoxtral_realtime
Layers3626
Native context32,768131,072
Mixture of expertsnono
Quantizations published38
Smallest quantization3.28 GiB2.35 GiB
Q4_K_M2.64 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextARK-ASR-3BVoxtral-Mini-4B-Realtime-2602Ratio
4,0960.14 GiB
8,1920.28 GiB
16,3840.56 GiB
32,7681.13 GiB
65,5362.25 GiB
131,0724.50 GiB