macos `amd64` compatibility fixes

llm/ggml_llama.go

@@ -45,6 +45,7 @@ func osPath(llamaPath string) string {
 	if runtime.GOOS == "windows" {
 		return path.Join(llamaPath, "Release")
 	}
+
 	return llamaPath
 }
 
@@ -68,7 +69,9 @@ func initGGML() {
 		case "windows":
 			files = []string{"server.exe"}
 		case "darwin":
-			files = append(files, "ggml-metal.metal")
+			if llamaPath == ggmlGPU {
+				files = append(files, "ggml-metal.metal")
+			}
 		}
 
 		for _, f := range files {

llm/llama.cpp/generate_darwin_amd64.go

@@ -6,5 +6,5 @@ package llm
 //go:generate git -C ggml apply ../ggml_patch/0002-34B-model-support.patch
 //go:generate git -C ggml apply ../ggml_patch/0003-metal-fix-synchronization-in-new-matrix-multiplicati.patch
 //go:generate git -C ggml apply ../ggml_patch/0004-metal-add-missing-barriers-for-mul-mat-2699.patch
-//go:generate cmake --fresh -S ggml -B ggml/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64
+//go:generate cmake --fresh -S ggml -B ggml/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=10.11
 //go:generate cmake --build ggml/build/cpu --target server --config Release

llm/llama.cpp/generate_darwin_arm64.go

@@ -6,5 +6,5 @@ package llm
 //go:generate git -C ggml apply ../ggml_patch/0002-34B-model-support.patch
 //go:generate git -C ggml apply ../ggml_patch/0003-metal-fix-synchronization-in-new-matrix-multiplicati.patch
 //go:generate git -C ggml apply ../ggml_patch/0004-metal-add-missing-barriers-for-mul-mat-2699.patch
-//go:generate cmake --fresh -S ggml -B ggml/build/gpu -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64
+//go:generate cmake --fresh -S ggml -B ggml/build/gpu -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
 //go:generate cmake --build ggml/build/gpu --target server --config Release

scripts/build_darwin.sh

@@ -8,6 +8,7 @@ GO_LDFLAGS="$GO_LDFLAGS -X github.com/jmorganca/ollama/server.mode=release"
 # build universal binary
 GOARCH=arm64 go generate ./...
 GOARCH=arm64 go build -ldflags "$GO_LDFLAGS" -o dist/ollama-darwin-arm64
+rm -rf llm/llama.cpp/ggml/build/*/bin
 GOARCH=amd64 go generate ./...
 GOARCH=amd64 go build -ldflags "$GO_LDFLAGS" -o dist/ollama-darwin-amd64
 lipo -create -output dist/ollama dist/ollama-darwin-arm64 dist/ollama-darwin-amd64