Is gemma-2-2b-it-abliterated or Llama-3.1-8B-Instruct smaller?

gemma-2-2b-it-abliterated is smaller at Q4_K_M: 1.59 GiB against 4.58 GiB.

Which uses less memory at long context?

gemma-2-2b-it-abliterated — its KV cache at 32K context is 2.2x smaller, because of how its attention layers cache.

Model comparison

gemma-2-2b-it-abliterated vs Llama-3.1-8B-Instruct

At Q4_K_M, gemma-2-2b-it-abliterated is the smaller download — 1,708,582,784 bytes against 4,920,739,168. At long context the gap widens: gemma-2-2b-it-abliterated's KV cache at 32K is 2.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	gemma-2-2b-it-abliterated	Llama-3.1-8B-Instruct
Parameters	2.6B	8.0B
Architecture	gemma2	llama
Layers	26	32
Native context	8,192	131,072
Mixture of experts	no	no
Quantizations published	29	45
Smallest quantization	1.15 GiB	2.02 GiB
Q4_K_M	1.59 GiB	4.58 GiB
Licence	gemma	llama3.1

KV cache by context

the term that decides long-context viability

Context	gemma-2-2b-it-abliterated	Llama-3.1-8B-Instruct	Ratio
4,096	0.41 GiB	0.50 GiB	1.23×
8,192	0.63 GiB	1.00 GiB	1.58×
16,384	1.04 GiB	2.00 GiB	1.92×
32,768	1.85 GiB	4.00 GiB	2.16×
65,536	3.48 GiB	8.00 GiB	2.30×
131,072	6.73 GiB	16.00 GiB	2.38×