--- a/python/mlx/nn/layers/normalization.py
+++ b/python/mlx/nn/layers/normalization.py
@@ -257,9 +257,14 @@
         x = x.reshape(batch, -1, num_groups)
 
         # Normalize
-        means = mx.mean(x, axis=1, keepdims=True)
-        var = mx.var(x, axis=1, keepdims=True)
-        x = (x - means) * mx.rsqrt(var + self.eps)
+        if x.size and x.dtype in (mx.float16, mx.bfloat16):
+            x = mx.fast.layer_norm(
+                x.transpose(0, 2, 1), eps=self.eps, weight=None, bias=None
+            ).transpose(0, 2, 1)
+        else:
+            means = mx.mean(x, axis=1, keepdims=True)
+            var = mx.var(x, axis=1, keepdims=True)
+            x = (x - means) * mx.rsqrt(var + self.eps)
         x = x.reshape(batch, *rest, dims)
 
         return x