Built llama.cpp from source.
Started llama-server with Gemma model.
Tested model output and found garbage tokens.
Deleted llama.cpp source and model cache.
