Is Gemma-3-4b-it-Uncensored-DBL-X or Llama-3.1-8B-Instruct smaller?

Gemma-3-4b-it-Uncensored-DBL-X is smaller at Q4_K_M: 2.52 GiB against 4.58 GiB.

Which uses less memory at long context?

Gemma-3-4b-it-Uncensored-DBL-X — its KV cache at 32K context is 4.3x smaller, because of how its attention layers cache.

Model comparison

Gemma-3-4b-it-Uncensored-DBL-X vs Llama-3.1-8B-Instruct

At Q4_K_M, Gemma-3-4b-it-Uncensored-DBL-X is the smaller download — 2,709,839,808 bytes against 4,920,739,168. At long context the gap widens: Gemma-3-4b-it-Uncensored-DBL-X's KV cache at 32K is 4.3× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Gemma-3-4b-it-Uncensored-DBL-X	Llama-3.1-8B-Instruct
Parameters	4.7B	8.0B
Architecture	gemma3	llama
Layers	38	32
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	39	45
Smallest quantization	1.13 GiB	2.02 GiB
Q4_K_M	2.52 GiB	4.58 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	Gemma-3-4b-it-Uncensored-DBL-X	Llama-3.1-8B-Instruct	Ratio
4,096	0.28 GiB	0.50 GiB	1.78×
8,192	0.38 GiB	1.00 GiB	2.67×
16,384	0.56 GiB	2.00 GiB	3.56×
32,768	0.94 GiB	4.00 GiB	4.27×
65,536	1.69 GiB	8.00 GiB	4.74×
131,072	3.19 GiB	16.00 GiB	5.02×