YUAN runs GPT-OSS on NVIDIA Jetson.

YUAN has launched a GPT-OSS large language model configuration for NVIDIA Jetson. The Pandora toolkit runs inference on the edge platform so application data can remain on site.

Memory integrated with the Jetson module

NVIDIA Jetson modules include validated DRAM on the module. This removes a separate memory component from the carrier-board design and reduces the amount of memory sourcing and compatibility work required for a project.

What it does in practice

YUAN demonstrates two applications. A voice assistant handles visitor registration and inquiries. A video-analysis workflow uses crowd-density and dwell-time data to produce space-usage reports.

Deployment and module options

YUAN reports optimized inference of up to 16.2 tokens per second for the demonstrated configuration. The Pandora toolkit can be deployed through a web interface or terminal and supports Jetson Orin Nano, Orin NX, and AGX Orin options. Teams should verify model version, quantization, prompt length, and output settings when comparing throughput.