Model comparison

Tini-Cybersec-8B-A1B vs gemma-4-12B-it-qat-q4_0-unquantized

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Tini-Cybersec-8B-A1B's KV cache at 32K is 6.6× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Tini-Cybersec-8B-A1Bgemma-4-12B-it-qat-q4_0-unquantized
Parameters8.5B12.0B
Architecturelfm2moegemma4
Layers2448
Native context128,000262,144
Mixture of expertsyes, 32 expertsno
Quantizations published352
Smallest quantization2.71 GiB6.50 GiB
Q4_K_M4.80 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextTini-Cybersec-8B-A1Bgemma-4-12B-it-qat-q4_0-unquantizedRatio
4,0960.05 GiB0.72 GiB15.32×
8,1920.09 GiB0.97 GiB10.33×
16,3840.19 GiB1.47 GiB7.83×
32,7680.38 GiB2.47 GiB6.58×
65,5360.75 GiB4.47 GiB5.96×
131,0721.50 GiB8.47 GiB5.65×
Tini-Cybersec-8B-A1B vs gemma-4-12B-it-qat-q4_0-unquantized — size, memory and hardware fit — ossmodeldb