Is gemma-4-12B-it-qat-q4_0-unquantized or Qwen2.5-Coder-7B-Instruct-Uncensored smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

Qwen2.5-Coder-7B-Instruct-Uncensored — its KV cache at 32K context is 1.4x smaller, because of how its attention layers cache.

Model comparison

gemma-4-12B-it-qat-q4_0-unquantized vs Qwen2.5-Coder-7B-Instruct-Uncensored

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen2.5-Coder-7B-Instruct-Uncensored's KV cache at 32K is 1.4× 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	Qwen2.5-Coder-7B-Instruct-Uncensored
Parameters	12.0B	7.6B
Architecture	gemma4	qwen2
Layers	48	28
Native context	262,144	32,768
Mixture of experts	no	no
Quantizations published	2	8
Smallest quantization	6.50 GiB	2.81 GiB
Q4_K_M	—	4.36 GiB
Licence	apache-2.0	mit

KV cache by context

the term that decides long-context viability

Context	gemma-4-12B-it-qat-q4_0-unquantized	Qwen2.5-Coder-7B-Instruct-Uncensored	Ratio
4,096	0.72 GiB	0.22 GiB	3.29×
8,192	0.97 GiB	0.44 GiB	2.21×
16,384	1.47 GiB	0.88 GiB	1.68×
32,768	2.47 GiB	1.75 GiB	1.41×
65,536	4.47 GiB	3.50 GiB	1.28×
131,072	8.47 GiB	7.00 GiB	1.21×