Is gemma-4-12B-it-qat-q4_0-unquantized or Llama-3.1-Nemotron-70B-Instruct-HF 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 4.1x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs Llama-3.1-Nemotron-70B-Instruct-HF

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 4.1× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gemma-4-12B-it-qat-q4_0-unquantized	Llama-3.1-Nemotron-70B-Instruct-HF
Parameters	12.0B	70.6B
Architecture	gemma4	llama
Layers	48	80
Native context	262,144	131,072
Mixture of experts	no	no
Quantizations published	2	47
Smallest quantization	6.50 GiB	14.29 GiB
Q4_K_M	—	39.60 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	gemma-4-12B-it-qat-q4_0-unquantized	Llama-3.1-Nemotron-70B-Instruct-HF	Ratio
4,096	0.72 GiB	1.25 GiB	1.74×
8,192	0.97 GiB	2.50 GiB	2.58×
16,384	1.47 GiB	5.00 GiB	3.40×
32,768	2.47 GiB	10.00 GiB	4.05×
65,536	4.47 GiB	20.00 GiB	4.48×
131,072	8.47 GiB	40.00 GiB	4.72×