You need the newest llama.cpp and if you have an amd card and recently updated the drivers, roll them back. Most people complaining are using ROCm.

I assure you gemma 3 works fine in LM studio. Gguf and MLx are available.