Is gemma-4-12B-it-qat-q4_0-unquantized or Llama-2-7b-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 6.5x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs Llama-2-7b-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 6.5× 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-2-7b-hf
Parameters	12.0B	6.7B
Architecture	gemma4	llama
Layers	48	32
Native context	262,144	4,096
Mixture of experts	no	no
Quantizations published	2	12
Smallest quantization	6.50 GiB	2.63 GiB
Q4_K_M	—	3.80 GiB
Licence	apache-2.0	llama2

KV cache by context

the term that decides long-context viability

Context	gemma-4-12B-it-qat-q4_0-unquantized	Llama-2-7b-hf	Ratio
4,096	0.72 GiB	2.00 GiB	2.78×
8,192	0.97 GiB	4.00 GiB	4.13×
16,384	1.47 GiB	8.00 GiB	5.45×
32,768	2.47 GiB	16.00 GiB	6.48×
65,536	4.47 GiB	32.00 GiB	7.16×
131,072	8.47 GiB	64.00 GiB	7.56×