Is LFM2-VL-3B or Qwen2.5-VL-7B-Instruct smaller?

LFM2-VL-3B is smaller at Q4_K_M: 1.46 GiB against 4.36 GiB.

Which uses less memory at long context?

LFM2-VL-3B — its KV cache at 32K context is 3.5x smaller, because of how its attention layers cache.

Model comparison

LFM2-VL-3B vs Qwen2.5-VL-7B-Instruct

At Q4_K_M, LFM2-VL-3B is the smaller download — 1,563,668,256 bytes against 4,683,072,032. At long context the gap widens: LFM2-VL-3B's KV cache at 32K is 3.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	LFM2-VL-3B	Qwen2.5-VL-7B-Instruct
Parameters	3.0B	8.3B
Architecture	lfm2	qwen2vl
Layers	30	28
Native context	128,000	128,000
Mixture of experts	no	no
Quantizations published	7	50
Smallest quantization	1.38 GiB	1.93 GiB
Q4_K_M	1.46 GiB	4.36 GiB
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	LFM2-VL-3B	Qwen2.5-VL-7B-Instruct	Ratio
4,096	0.06 GiB	0.22 GiB	3.50×
8,192	0.13 GiB	0.44 GiB	3.50×
16,384	0.25 GiB	0.88 GiB	3.50×
32,768	0.50 GiB	1.75 GiB	3.50×
65,536	1.00 GiB	3.50 GiB	3.50×
131,072	2.00 GiB	7.00 GiB	3.50×