I saw this video about Ubuntu Inference Snaps that seemed interesting:
Video: https://youtu.be/0CYm-KCw7yY?si=S5CqQADktOy85Ea8&t=421
Docs: Get started with Inference Snaps using Qwen VL - Inference Snaps documentation
The idea is that you can install a model with snap install qwen-vl and it’ll automatically get the “right” one for your machine. Currently they have:
- DeepSeek R1
- Gemma 3
- Nemotron 3 Nano
- Qwen VL
The idea sounds interesting - it’d be nice to be able to try out a new model without first spending a day trying to figure out the best version and what magic flags you need to run it.
Curious if anyone has tried it (I think it might still be beta?). Given that he specifically mentions the Spark in the video, I assume it should work here (and I hope/assume it will be optimized for the hardware).