Is NVIDIA-Nemotron-3-Nano-4B-FP8 or gemma-4-12B-it-qat-q4_0-unquantized smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

gemma-4-12B-it-qat-q4_0-unquantized — its KV cache at 32K context is 2.1x smaller, because of how its attention layers cache.

Model comparison

NVIDIA-Nemotron-3-Nano-4B-FP8 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 2.1× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	NVIDIA-Nemotron-3-Nano-4B-FP8	gemma-4-12B-it-qat-q4_0-unquantized
Parameters	4.0B	12.0B
Architecture	nemotron_h	gemma4
Layers	42	48
Native context	262,144	262,144
Mixture of experts	no	no
Quantizations published	1	2
Smallest quantization	2.64 GiB	6.50 GiB
Q4_K_M	2.64 GiB	—
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	NVIDIA-Nemotron-3-Nano-4B-FP8	gemma-4-12B-it-qat-q4_0-unquantized	Ratio
4,096	0.66 GiB	0.72 GiB	1.10×
8,192	1.31 GiB	0.97 GiB	1.35×
16,384	2.63 GiB	1.47 GiB	1.79×
32,768	5.25 GiB	2.47 GiB	2.13×
65,536	10.50 GiB	4.47 GiB	2.35×
131,072	21.00 GiB	8.47 GiB	2.48×