DeepSeek R1 distilled model running locally on AI Base Station based on Nvidia Jetson Orin Nano SBC with only 16GB unified memory?

[Hedda]

One more discussion related to all and any coming AI Base Station models comming from FutureProofHomes and others.

Has anyone else tried running Deepseek-R1 distill model locally on low-end variants of Nvidia Jetson Nano series of single-board computers?

Spotted several people running distilled model of Deepseek R-1 LLM on the Nvidia Jetson Orin Nano SBC with only 16GB unified memory:

https://www.youtube.com/watch?v=e-EG3B5Uj78

https://www.youtube.com/watch?v=YnoeWXh3-q0

https://www.nextbigfuture.com/2025/01/tiny-deepseek-1-5b-models-run-on-99-nvidia-jetson-nano.html

https://blogs.nvidia.com/blog/deepseek-r1-nim-microservice/

If setup as a local local Ollama server then can use this from Home Assistant's Assist and other applications via the Ollama integration:

https://www.home-assistant.io/integrations/ollama/

https://www.home-assistant.io/voice_control/assist_create_open_ai_personality/

Tip on another cool custom Home Assistant integration is this AI Automation Suggester:

https://github.com/ITSpecialist111/ai_automation_suggester

PS: Nvidia's JetsonNano series features a unified physical memory architecture between the GPU and CPU, (similar in physical memory architecture to Apple M-Series computers as well as the new AMD Ryzen™ Al Max+ 395 which also share same memory between CPU and GPU).