Is G9v3-3B or Llama-3.1-8B-Instruct smaller?

G9v3-3B is smaller at Q4_K_M: 1.72 GiB against 4.58 GiB.

Which uses less memory at long context?

G9v3-3B — its KV cache at 32K context is 2.5x smaller, because of how its attention layers cache.

Model comparison

G9v3-3B vs Llama-3.1-8B-Instruct

At Q4_K_M, G9v3-3B is the smaller download — 1,843,804,448 bytes against 4,920,739,168. At long context the gap widens: G9v3-3B's KV cache at 32K is 2.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	G9v3-3B	Llama-3.1-8B-Instruct
Parameters	3.0B	8.0B
Architecture	llama	llama
Layers	52	32
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	40	45
Smallest quantization	1.15 GiB	2.02 GiB
Q4_K_M	1.72 GiB	4.58 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	G9v3-3B	Llama-3.1-8B-Instruct	Ratio
4,096	0.20 GiB	0.50 GiB	2.46×
8,192	0.41 GiB	1.00 GiB	2.46×
16,384	0.81 GiB	2.00 GiB	2.46×
32,768	1.63 GiB	4.00 GiB	2.46×
65,536	3.25 GiB	8.00 GiB	2.46×
131,072	6.50 GiB	16.00 GiB	2.46×