GRL or Grillcheese Research Laboratory started with an experimental project called Grillcheese AI. The goal was to see how a single individual could achieve what large teams of PHD were achieving in the Deep learning space by combining 25 years of human coding experience with the collaboration of diverse AI. Claude Code (MAX), Gemini 2.5 and 3 pro and Openrouter were used as AI collaborators to test various experiments.

IMPORTANT: The experiments were planned in advanced and not done via vibe-coding, structure was put in place with serious scientific protocols.

I had a problem. I was only having an old AMD RX 6750 XT with 12GB VRAM. Plenty for every day stuff. But not really hot for deep learning. No torch, No CUDA, No AMD ROCM. Very slow. Pytorch was basically a big numpy machine for me. So I decided to build a framework that uses vulkan to be able to do what pytorch was doing and more on my humble setup. I wasn't sure it would work at first but I tried and well Grilly was born!

Enjoy the framework, check our latest experiment, open issues, discuss about it or Join me into making deep learning more accessible and green !