Model comparison

SenseVoiceSmall vs whisper-large-v3-turbo

At Q4_K_M, SenseVoiceSmall is the smaller download — 145,738,304 bytes against 536,069,728.

From the file· summed bytes, KV per layer

Side by side

SenseVoiceSmallwhisper-large-v3-turbo
Parameters234M809M
Architecturesensevoicewhisper
Layers32
Native context
Mixture of expertsnono
Quantizations published931
Smallest quantization0.13 GiB0.27 GiB
Q4_K_M0.14 GiB0.50 GiB
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
ContextSenseVoiceSmallwhisper-large-v3-turboRatio
4,096
8,192
16,384
32,768
65,536
131,072