Model comparison

gemma-4-E4B-it-assistant vs tinygemma3_cifar

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gemma-4-E4B-it-assistant's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

gemma-4-E4B-it-assistanttinygemma3_cifar
Parameters79M39M
Architecturegemma4-assistantgemma3
Layers48
Native context131,072131,072
Mixture of expertsnono
Quantizations published91
Smallest quantization0.07 GiB0.04 GiB
Q4_K_M0.07 GiB
Licenceapache-2.0wtfpl

KV cache by context

the term that decides long-context viability
Contextgemma-4-E4B-it-assistanttinygemma3_cifarRatio
4,0960.01 GiB0.06 GiB4.57×
8,1920.02 GiB0.08 GiB3.59×
16,3840.04 GiB0.09 GiB2.50×
32,7680.07 GiB0.12 GiB1.81×
65,5360.13 GiB0.19 GiB1.43×
131,0720.26 GiB0.31 GiB1.22×