Adapted rocm support to cgo based llama.cpp

llm/shim_darwin.go

@@ -0,0 +1,18 @@
+package llm
+
+import (
+	"fmt"
+
+	"github.com/jmorganca/ollama/api"
+)
+
+// no-op stubs for mac
+
+func newRocmShimExtServer(model string, adapters, projectors []string, numLayers int64, opts api.Options) (extServer, error) {
+	// should never happen...
+	return nil, fmt.Errorf("ROCM GPUs not supported on Mac")
+}
+
+func nativeInit(workDir string) error {
+	return nil
+}