ggml-metal.metal must be in the same directory as the ollama binary otherwise llama.cpp will not be able to find it and load it. 1. go generate llama/llama_metal.go 2. go build . 3. ./ollama serve
