Is SmolVLM-Instruct or Llama-3.1-8B-Instruct smaller?

SmolVLM-Instruct is smaller at Q4_K_M: 1.04 GiB against 4.58 GiB.

Which uses less memory at long context?

Llama-3.1-8B-Instruct — its KV cache at 32K context is 1.5x smaller, because of how its attention layers cache.

Model comparison

SmolVLM-Instruct vs Llama-3.1-8B-Instruct

At Q4_K_M, SmolVLM-Instruct is the smaller download — 1,112,242,368 bytes against 4,920,739,168. At long context the gap widens: Llama-3.1-8B-Instruct's KV cache at 32K is 1.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	SmolVLM-Instruct	Llama-3.1-8B-Instruct
Parameters	2.2B	8.0B
Architecture	llama	llama
Layers	24	32
Native context	16,384	131,072
Mixture of experts	no	no
Quantizations published	3	45
Smallest quantization	1.04 GiB	2.02 GiB
Q4_K_M	1.04 GiB	4.58 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	SmolVLM-Instruct	Llama-3.1-8B-Instruct	Ratio
4,096	0.75 GiB	0.50 GiB	1.50×
8,192	1.50 GiB	1.00 GiB	1.50×
16,384	3.00 GiB	2.00 GiB	1.50×
32,768	6.00 GiB	4.00 GiB	1.50×
65,536	12.00 GiB	8.00 GiB	1.50×
131,072	24.00 GiB	16.00 GiB	1.50×