Model comparison

whisper-small.en vs whisper-large-v3-turbo

At Q4_K_M, whisper-small.en is the smaller download — 171,553,856 bytes against 536,069,728.

From the file· summed bytes, KV per layer

Side by side

whisper-small.enwhisper-large-v3-turbo
Parameters242M809M
Architecturewhisperwhisper
Layers1232
Native context
Mixture of expertsnono
Quantizations published631
Smallest quantization0.16 GiB0.27 GiB
Q4_K_M0.16 GiB0.50 GiB
Licenceapache-2.0apache-2.0

KV cache by context

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