Is Phi-3.5-mini-instruct or Llama-3.2-1B-Instruct smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

Llama-3.2-1B-Instruct — its KV cache at 32K context is 12.0x smaller, because of how its attention layers cache.

Model comparison

Phi-3.5-mini-instruct vs Llama-3.2-1B-Instruct

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Llama-3.2-1B-Instruct's KV cache at 32K is 12.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Phi-3.5-mini-instruct	Llama-3.2-1B-Instruct
Parameters	3.8B	1.2B
Architecture	llama	llama
Layers	32	16
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	24	39
Smallest quantization	0.82 GiB	0.39 GiB
Q4_K_M	—	0.75 GiB
Licence	mit	—

KV cache by context

the term that decides long-context viability

Context	Phi-3.5-mini-instruct	Llama-3.2-1B-Instruct	Ratio
4,096	1.50 GiB	0.13 GiB	12.00×
8,192	3.00 GiB	0.25 GiB	12.00×
16,384	6.00 GiB	0.50 GiB	12.00×
32,768	12.00 GiB	1.00 GiB	12.00×
65,536	24.00 GiB	2.00 GiB	12.00×
131,072	48.00 GiB	4.00 GiB	12.00×