Is Firefly-v4 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?

Firefly-v4 — its KV cache at 32K context is 10.0x smaller, because of how its attention layers cache.

Model comparison

Firefly-v4 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: Firefly-v4's KV cache at 32K is 10.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Firefly-v4	gemma-4-12B-it-qat-q4_0-unquantized
Parameters	5.1B	12.0B
Architecture	gemma4	gemma4
Layers	35	48
Native context	131,072	262,144
Mixture of experts	no	no
Quantizations published	27	2
Smallest quantization	2.78 GiB	6.50 GiB
Q4_K_M	3.19 GiB	—
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	Firefly-v4	gemma-4-12B-it-qat-q4_0-unquantized	Ratio
4,096	0.05 GiB	0.72 GiB	13.14×
8,192	0.08 GiB	0.97 GiB	11.81×
16,384	0.14 GiB	1.47 GiB	10.74×
32,768	0.25 GiB	2.47 GiB	10.03×
65,536	0.46 GiB	4.47 GiB	9.61×
131,072	0.90 GiB	8.47 GiB	9.39×