The project
Gemma 4 Local Lab is an environment for testing local Gemma models with a real interface, FastAPI backend, React frontend, local speech and several execution paths.
Technical angle
The repo documents BF16 through Transformers, Q8/Q4 through llama.cpp and NVFP4 through vLLM on WSL, with preload scripts, shared cache and RTX 5090 benchmarks.
What it proves
It shows local AI infrastructure skill: install, measure, compare and make heavy models usable on a real workstation.