diff --git a/llm/llm.go b/llm/llm.go
index b537865e..020e3c2f 100644
--- a/llm/llm.go
+++ b/llm/llm.go
@@ -2,6 +2,7 @@ package llm
 
 import (
 	"fmt"
+	"log"
 	"os"
 
 	"github.com/jmorganca/ollama/api"
@@ -31,6 +32,16 @@ func New(model string, opts api.Options) (LLM, error) {
 		return nil, err
 	}
 
+	switch ggml.FileType {
+	case FileTypeF32, FileTypeF16, FileTypeQ5_0, FileTypeQ5_1, FileTypeQ8_0:
+		if opts.NumGPU != 0 {
+			// Q5_0, Q5_1, and Q8_0 do not support Metal API and will
+			// cause the runner to segmentation fault so disable GPU
+			log.Printf("WARNING: GPU disabled for F32, F16, Q5_0, Q5_1, and Q8_0")
+			opts.NumGPU = 0
+		}
+	}
+
 	switch ggml.ModelFamily {
 	case ModelFamilyLlama:
 		return newLlama(model, opts)