Model comparison

tada-3b-ml vs s2-pro

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

From the file· summed bytes, KV per layer

Side by side

tada-3b-mls2-pro
Parameters4.2B4.6B
Architecturewav2vec2fish-speech
Layers2836
Native context131,072
Mixture of expertsnoyes, 1 experts
Quantizations published37
Smallest quantization5.20 GiB2.40 GiB
Q4_K_M3.32 GiB
Licencellama3.2other

KV cache by context

the term that decides long-context viability
Contexttada-3b-mls2-proRatio
4,0960.44 GiB
8,1920.88 GiB
16,3841.75 GiB
32,7683.50 GiB
65,5367.00 GiB
131,07214.00 GiB