diff --git a/llms/hf_llm/convert.py b/llms/hf_llm/convert.py
index 0704b24b..f93d01c3 100644
--- a/llms/hf_llm/convert.py
+++ b/llms/hf_llm/convert.py
@@ -154,12 +154,12 @@ if __name__ == "__main__":
 
     print("[INFO] Loading")
     weights, config, tokenizer = fetch_from_hub(args.hf_path)
+
+    dtype = mx.float16 if args.quantize else getattr(mx, args.dtype)
+    weights = {k: v.astype(dtype) for k, v in weights.items()}
     if args.quantize:
         print("[INFO] Quantizing")
         weights, config = quantize(weights, config, args)
-    if not args.quantize:
-        dtype = getattr(mx, args.dtype)
-        weights = {k: v.astype(dtype) for k, v in weights.items()}
 
     mlx_path = Path(args.mlx_path)
     mlx_path.mkdir(parents=True, exist_ok=True)