Model comparison

parakeet-tdt-0.6b-v3 vs whisper-medium

At Q4_K_M, parakeet-tdt-0.6b-v3 is the smaller download — 485,425,504 bytes against 504,102,848.

From the file· summed bytes, KV per layer

Side by side

parakeet-tdt-0.6b-v3whisper-medium
Parameters627M764M
Architectureparakeetwhisper
Layers24
Native context
Mixture of expertsnono
Quantizations published1019
Smallest quantization0.39 GiB0.25 GiB
Q4_K_M0.45 GiB0.47 GiB
Licencecc-by-4.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextparakeet-tdt-0.6b-v3whisper-mediumRatio
4,096
8,192
16,384
32,768
65,536
131,072