Is LFM2-350M or SmolLM2-135M-Instruct smaller?

SmolLM2-135M-Instruct is smaller at Q4_K_M: 0.10 GiB against 0.21 GiB.

Which uses less memory at long context?

SmolLM2-135M-Instruct — its KV cache at 32K context is 1.4x smaller, because of how its attention layers cache.

Model comparison

LFM2-350M vs SmolLM2-135M-Instruct

At Q4_K_M, SmolLM2-135M-Instruct is the smaller download — 105,454,144 bytes against 229,309,152. At long context the gap widens: SmolLM2-135M-Instruct's KV cache at 32K is 1.4× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	LFM2-350M	SmolLM2-135M-Instruct
Parameters	354M	135M
Architecture	lfm2	llama
Layers	16	30
Native context	128,000	8,192
Mixture of experts	no	no
Quantizations published	19	41
Smallest quantization	0.15 GiB	0.08 GiB
Q4_K_M	0.21 GiB	0.10 GiB
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	LFM2-350M	SmolLM2-135M-Instruct	Ratio
4,096	0.13 GiB	0.09 GiB	1.42×
8,192	0.25 GiB	0.18 GiB	1.42×
16,384	0.50 GiB	0.35 GiB	1.42×
32,768	1.00 GiB	0.70 GiB	1.42×
65,536	2.00 GiB	1.41 GiB	1.42×
131,072	4.00 GiB	2.81 GiB	1.42×