LFM2.5-VL-3B on DGX Spark — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	LFM2.5-VL-3B
Parameters	3.1B
Device	DGX Spark
Quantization	BF16
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	64
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.18	34.3	≤3 s
2	0.25	39.2	≤3 s
4	0.29	37.6	≤3 s
8	0.43	34.5	≤3 s
16	0.92	30.9	≤3 s
32	1.53	23.9	≤3 s
64	2.40	15.7	≤3 s
128	5.22	9.7	>3 s
256	19.7	8.1	>3 s
288	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.18	34.3	≤3 s
2	0.31	34.1	≤3 s
4	0.61	34.0	≤3 s
8	1.20	33.7	≤3 s
16	2.42	33.2	≤3 s
32	4.98	36.6	>3 s
64	—	—	Failed

LFM2.5-VL-3B (3.1B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.18 s, then writes 34.3 tok/s.

With 32 images per camera at 720p, the answer starts after 4.98 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 64 cameras at once.