Is Llama-2-7B-32K-Instruct or Qwen3-4B smaller?

Qwen3-4B is smaller at Q4_K_M: 2.33 GiB against 3.80 GiB.

Which uses less memory at long context?

Qwen3-4B — its KV cache at 32K context is 3.6x smaller, because of how its attention layers cache.

Model comparison

Llama-2-7B-32K-Instruct vs Qwen3-4B

At Q4_K_M, Qwen3-4B is the smaller download — 2,497,280,256 bytes against 4,081,004,288. At long context the gap widens: Qwen3-4B's KV cache at 32K is 3.6× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Llama-2-7B-32K-Instruct	Qwen3-4B
Parameters	6.7B	4.0B
Architecture	llama	qwen3
Layers	32	36
Native context	32,768	40,960
Mixture of experts	no	no
Quantizations published	36	29
Smallest quantization	1.42 GiB	1.01 GiB
Q4_K_M	3.80 GiB	2.33 GiB
Licence	llama2	—

KV cache by context

the term that decides long-context viability

Context	Llama-2-7B-32K-Instruct	Qwen3-4B	Ratio
4,096	2.00 GiB	0.56 GiB	3.56×
8,192	4.00 GiB	1.13 GiB	3.56×
16,384	8.00 GiB	2.25 GiB	3.56×
32,768	16.00 GiB	4.50 GiB	3.56×
65,536	32.00 GiB	9.00 GiB	3.56×
131,072	64.00 GiB	18.00 GiB	3.56×