diff --git a/docs/src/dev/custom_metal_kernels.rst b/docs/src/dev/custom_metal_kernels.rst
index 873b1e544..1febe960a 100644
--- a/docs/src/dev/custom_metal_kernels.rst
+++ b/docs/src/dev/custom_metal_kernels.rst
@@ -128,6 +128,7 @@ relying on a copy from ``ensure_row_contiguous``:
       input_names=["inp"],
       output_names=["out"],
       source=source
+      ensure_row_contiguous=False,
   )
 
   def exp_elementwise(a: mx.array):
@@ -138,7 +139,6 @@ relying on a copy from ``ensure_row_contiguous``:
           threadgroup=(256, 1, 1),
           output_shapes=[a.shape],
           output_dtypes=[a.dtype],
-          ensure_row_contiguous=False,
       )
       return outputs[0]
 
diff --git a/mlx/ops.cpp b/mlx/ops.cpp
index 14e135deb..c8583c72f 100644
--- a/mlx/ops.cpp
+++ b/mlx/ops.cpp
@@ -2971,7 +2971,7 @@ array gather(
   }
   for (auto& x : indices) {
     if (x.dtype() == bool_) {
-      throw("[Gather] Boolean indices not supported.");
+      throw std::invalid_argument("[Gather] Boolean indices not supported.");
     }
   }
 
diff --git a/python/tests/test_fast.py b/python/tests/test_fast.py
index 13c65de99..f79a62a15 100644
--- a/python/tests/test_fast.py
+++ b/python/tests/test_fast.py
@@ -639,12 +639,12 @@ class TestFast(mlx_tests.MLXTestCase):
             ],
             grid=(6, 1, 1),
             threadgroup=(2, 1, 1),
-            output_shapes=[(2, 2), (3, 2)],
+            output_shapes=[(3, 2), (3, 2)],
             output_dtypes=[mx.float32, mx.int32],
             stream=mx.gpu,
         )
 
-        self.assertTrue(mx.allclose(out[0], mx.full((2, 2), 14.0484)))
+        self.assertTrue(mx.allclose(out[0], mx.full((3, 2), 14.0484)))
         self.assertTrue(mx.allclose(out[1], mx.full((3, 2), -2, dtype=mx.int32)))
 
     @unittest.skipIf(not mx.metal.is_available(), "Metal is not available")