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

LocateAnything-3B is smaller at Q4_K_M: 1.96 GiB against 4.36 GiB.

Which uses less memory at long context?

LocateAnything-3B — its KV cache at 32K context is 1.6x smaller, because of how its attention layers cache.

Model comparison

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

At Q4_K_M, LocateAnything-3B is the smaller download — 2,107,054,176 bytes against 4,683,072,032. At long context the gap widens: LocateAnything-3B's KV cache at 32K is 1.6× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	LocateAnything-3B	Qwen2.5-VL-7B-Instruct
Parameters	3.8B	8.3B
Architecture	qwen2	qwen2vl
Layers	36	28
Native context	32,768	128,000
Mixture of experts	no	no
Quantizations published	17	50
Smallest quantization	1.28 GiB	1.93 GiB
Q4_K_M	1.96 GiB	4.36 GiB
Licence	other	apache-2.0

KV cache by context

the term that decides long-context viability

Context	LocateAnything-3B	Qwen2.5-VL-7B-Instruct	Ratio
4,096	0.14 GiB	0.22 GiB	1.56×
8,192	0.28 GiB	0.44 GiB	1.56×
16,384	0.56 GiB	0.88 GiB	1.56×
32,768	1.13 GiB	1.75 GiB	1.56×
65,536	2.25 GiB	3.50 GiB	1.56×
131,072	4.50 GiB	7.00 GiB	1.56×