Is NVIDIA-Nemotron-3-Nano-4B-FP8 or gemma-4-E4B-it smaller?

NVIDIA-Nemotron-3-Nano-4B-FP8 is smaller at Q4_K_M: 2.64 GiB against 4.64 GiB.

Which uses less memory at long context?

gemma-4-E4B-it — its KV cache at 32K context is 10.4x smaller, because of how its attention layers cache.

Model comparison

NVIDIA-Nemotron-3-Nano-4B-FP8 vs gemma-4-E4B-it

At Q4_K_M, NVIDIA-Nemotron-3-Nano-4B-FP8 is the smaller download — 2,837,072,864 bytes against 4,977,171,584. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 10.4× 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-E4B-it
Parameters	4.0B	8.0B
Architecture	nemotron_h	gemma4
Layers	42	42
Native context	262,144	131,072
Mixture of experts	no	no
Quantizations published	1	36
Smallest quantization	2.64 GiB	3.30 GiB
Q4_K_M	2.64 GiB	4.64 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-E4B-it	Ratio
4,096	0.66 GiB	0.12 GiB	5.33×
8,192	1.31 GiB	0.18 GiB	7.38×
16,384	2.63 GiB	0.29 GiB	9.14×
32,768	5.25 GiB	0.51 GiB	10.38×
65,536	10.50 GiB	0.94 GiB	11.13×
131,072	21.00 GiB	1.82 GiB	11.55×