Is LFM2-350M or Qwen2.5-0.5B-Instruct smaller?

LFM2-350M is smaller at Q4_K_M: 0.21 GiB against 0.37 GiB.

Which uses less memory at long context?

Qwen2.5-0.5B-Instruct — its KV cache at 32K context is 2.7x smaller, because of how its attention layers cache.

Model comparison

LFM2-350M vs Qwen2.5-0.5B-Instruct

At Q4_K_M, LFM2-350M is the smaller download — 229,309,152 bytes against 397,807,936. At long context the gap widens: Qwen2.5-0.5B-Instruct's KV cache at 32K is 2.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	LFM2-350M	Qwen2.5-0.5B-Instruct
Parameters	354M	494M
Architecture	lfm2	qwen2
Layers	16	24
Native context	128,000	32,768
Mixture of experts	no	no
Quantizations published	19	32
Smallest quantization	0.15 GiB	0.31 GiB
Q4_K_M	0.21 GiB	0.37 GiB
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	LFM2-350M	Qwen2.5-0.5B-Instruct	Ratio
4,096	0.13 GiB	0.05 GiB	2.67×
8,192	0.25 GiB	0.09 GiB	2.67×
16,384	0.50 GiB	0.19 GiB	2.67×
32,768	1.00 GiB	0.38 GiB	2.67×
65,536	2.00 GiB	0.75 GiB	2.67×
131,072	4.00 GiB	1.50 GiB	2.67×