Model comparison

gemma-4-12B-it vs gemma-4-E4B-it-ultra-uncensored-heretic

At Q4_K_M, gemma-4-E4B-it-ultra-uncensored-heretic is the smaller download — 5,335,286,336 bytes against 7,121,861,440. At long context the gap widens: gemma-4-E4B-it-ultra-uncensored-heretic's KV cache at 32K is 4.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

gemma-4-12B-itgemma-4-E4B-it-ultra-uncensored-heretic
Parameters12.0B8.0B
Architecturegemma4gemma4
Layers4842
Native context262,144131,072
Mixture of expertsnono
Quantizations published5924
Smallest quantization3.92 GiB4.10 GiB
Q4_K_M6.63 GiB4.97 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextgemma-4-12B-itgemma-4-E4B-it-ultra-uncensored-hereticRatio
4,0960.72 GiB0.12 GiB5.84×
8,1920.97 GiB0.18 GiB5.45×
16,3841.47 GiB0.29 GiB5.12×
32,7682.47 GiB0.51 GiB4.88×
65,5364.47 GiB0.94 GiB4.74×
131,0728.47 GiB1.82 GiB4.66×