Model comparison

AfriqueGemma-12B vs gemma-4-12B-it

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

From the file· summed bytes, KV per layer

Side by side

AfriqueGemma-12Bgemma-4-12B-it
Parameters12.2B12.0B
Architecturegemma3gemma4
Layers4848
Native context131,072262,144
Mixture of expertsnono
Quantizations published2459
Smallest quantization3.05 GiB3.92 GiB
Q4_K_M6.63 GiB
Licencecc-by-4.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextAfriqueGemma-12Bgemma-4-12B-itRatio
4,0960.72 GiB0.72 GiB1.00×
8,1920.97 GiB0.97 GiB1.00×
16,3841.47 GiB1.47 GiB1.00×
32,7682.47 GiB2.47 GiB1.00×
65,5364.47 GiB4.47 GiB1.00×
131,0728.47 GiB8.47 GiB1.00×