Is deepseek-llm-67b-base 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 4.8x smaller, because of how its attention layers cache.

Model comparison

deepseek-llm-67b-base 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 4.8× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	deepseek-llm-67b-base	gemma-4-12B-it-qat-q4_0-unquantized
Parameters	67.4B	12.0B
Architecture	llama	gemma4
Layers	95	48
Native context	4,096	262,144
Mixture of experts	no	no
Quantizations published	30	2
Smallest quantization	13.65 GiB	6.50 GiB
Q4_K_M	37.66 GiB	—
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	deepseek-llm-67b-base	gemma-4-12B-it-qat-q4_0-unquantized	Ratio
4,096	1.48 GiB	0.72 GiB	2.07×
8,192	2.97 GiB	0.97 GiB	3.06×
16,384	5.94 GiB	1.47 GiB	4.04×
32,768	11.88 GiB	2.47 GiB	4.81×
65,536	23.75 GiB	4.47 GiB	5.31×
131,072	47.50 GiB	8.47 GiB	5.61×