LensVLM-9B on DGX Spark — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	LensVLM-9B
Parameters	9.4B
Device	DGX Spark
Quantization	BF16
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	16
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.37	13.0	≤3 s
2	0.46	13.7	≤3 s
4	0.51	12.8	≤3 s
8	0.87	11.8	≤3 s
16	1.77	9.9	≤3 s
32	3.33	7.2	>3 s
64	6.42	4.8	>3 s
80	8.28	4.0	>3 s
88	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.37	13.0	≤3 s
2	0.66	12.9	≤3 s
4	1.25	13.1	≤3 s
8	2.49	13.0	≤3 s
16	5.04	12.6	>3 s
32	10.4	12.5	>3 s
64	22.2	11.8	>3 s
128	50.7	11.0	>3 s
256	—	—	Failed

LensVLM-9B (9.4B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.37 s, then writes 13 tok/s.

With 128 images per camera at 720p, the answer starts after 50.71 s. With one 2K image, after 0.66 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 16 cameras at once.