Is phi-2 or Llama-3.1-8B-Instruct smaller?

phi-2 is smaller at Q4_K_M: 1.62 GiB against 4.58 GiB.

Which uses less memory at long context?

Llama-3.1-8B-Instruct — its KV cache at 32K context is 2.5x smaller, because of how its attention layers cache.

Model comparison

phi-2 vs Llama-3.1-8B-Instruct

At Q4_K_M, phi-2 is the smaller download — 1,737,636,384 bytes against 4,920,739,168. At long context the gap widens: Llama-3.1-8B-Instruct'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

	phi-2	Llama-3.1-8B-Instruct
Parameters	2.8B	8.0B
Architecture	phi2	llama
Layers	32	32
Native context	2,048	131,072
Mixture of experts	no	no
Quantizations published	34	45
Smallest quantization	1.03 GiB	2.02 GiB
Q4_K_M	1.62 GiB	4.58 GiB
Licence	other	llama3.1

KV cache by context

the term that decides long-context viability

Context	phi-2	Llama-3.1-8B-Instruct	Ratio
4,096	1.25 GiB	0.50 GiB	2.50×
8,192	2.50 GiB	1.00 GiB	2.50×
16,384	5.00 GiB	2.00 GiB	2.50×
32,768	10.00 GiB	4.00 GiB	2.50×
65,536	20.00 GiB	8.00 GiB	2.50×
131,072	40.00 GiB	16.00 GiB	2.50×