Is gemma4-12B-it-DFlash or Llama-3.2-1B-Instruct smaller?

gemma4-12B-it-DFlash is smaller at Q4_K_M: 0.41 GiB against 0.75 GiB.

Which uses less memory at long context?

gemma4-12B-it-DFlash — its KV cache at 32K context is 6.1x smaller, because of how its attention layers cache.

Model comparison

gemma4-12B-it-DFlash vs Llama-3.2-1B-Instruct

At Q4_K_M, gemma4-12B-it-DFlash is the smaller download — 442,398,496 bytes against 807,690,688. At long context the gap widens: gemma4-12B-it-DFlash's KV cache at 32K is 6.1× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gemma4-12B-it-DFlash	Llama-3.2-1B-Instruct
Parameters	727M	1.2B
Architecture	dflash	llama
Layers	5	16
Native context	262,144	131,072
Mixture of experts	no	no
Quantizations published	6	39
Smallest quantization	0.38 GiB	0.39 GiB
Q4_K_M	0.41 GiB	0.75 GiB
Licence	apache-2.0	—

KV cache by context

the term that decides long-context viability

Context	gemma4-12B-it-DFlash	Llama-3.2-1B-Instruct	Ratio
4,096	0.05 GiB	0.13 GiB	2.29×
8,192	0.07 GiB	0.25 GiB	3.56×
16,384	0.10 GiB	0.50 GiB	4.92×
32,768	0.16 GiB	1.00 GiB	6.10×
65,536	0.29 GiB	2.00 GiB	6.92×
131,072	0.54 GiB	4.00 GiB	7.42×