~21 tok/s Gemma 4 on a Ryzen mini PC: llama.cpp, Vulkan, and the messy truth about local chat

· Dev.to

Read full story at source