Model comparison

llama2-22b-chat-wizard-uncensored 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: gemma-4-12B-it-qat-q4_0-unquantized's KV cache at 32K is 13.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

llama2-22b-chat-wizard-uncensoredgemma-4-12B-it-qat-q4_0-unquantized
Parameters21.8B12.0B
Architecturellamagemma4
Layers4048
Native context2,048262,144
Mixture of expertsnono
Quantizations published22
Smallest quantization7.56 GiB6.50 GiB
Q4_K_M
Licenceotherapache-2.0

KV cache by context

the term that decides long-context viability
Contextllama2-22b-chat-wizard-uncensoredgemma-4-12B-it-qat-q4_0-unquantizedRatio
4,0964.06 GiB0.72 GiB5.65×
8,1928.13 GiB0.97 GiB8.39×
16,38416.25 GiB1.47 GiB11.06×
32,76832.50 GiB2.47 GiB13.16×
65,53665.00 GiB4.47 GiB14.55×
131,072130.00 GiB8.47 GiB15.35×
llama2-22b-chat-wizard-uncensored vs gemma-4-12B-it-qat-q4_0-unquantized — size, memory and hardware fit — ossmodeldb