fix error detection by limiting model loading error parsing (#5472)

llm/status.go

@@ -25,7 +25,7 @@ var errorPrefixes = []string{
 	"CUDA error",
 	"cudaMalloc failed",
 	"\"ERR\"",
-	"architecture",
+	"error loading model",
 }
 
 func (w *StatusWriter) Write(b []byte) (int, error) {