diff --git a/python/tvm/topi/cuda/conv2d_transpose.py b/python/tvm/topi/cuda/conv2d_transpose.py
index 3d308474bc50b..6a614912a969b 100644
--- a/python/tvm/topi/cuda/conv2d_transpose.py
+++ b/python/tvm/topi/cuda/conv2d_transpose.py
@@ -59,7 +59,10 @@ def conv2d_transpose_nchw(cfg, data, kernel, stride, padding, out_dtype, output_
     stride_height, stride_width = stride
     outpad_height, outpad_width = output_padding
     assert outpad_height < stride_height and outpad_width < stride_width
-    assert inp_channels % groups == 0, f"input channels {inp_channels} must divide group size {groups}"
+
+    if inp_channels % groups != 0:
+        raise ValueError(f"input channels {inp_channels} must divide group size {groups}")
+
     cfg.stride = stride
     pad_top, pad_left, pad_bottom, pad_right = nn.get_pad_tuple(
         padding, (kernel_height, kernel_width)
@@ -112,14 +115,14 @@ def conv2d_transpose_nchw(cfg, data, kernel, stride, padding, out_dtype, output_
     data_out = te.compute(
         (batch, out_channels * groups, out_height, out_width),
         lambda b, c, h, w: te.sum(
-            data[
-                b,  c // out_channels * (inp_channels // groups) + dc, h + dh, w + dw
-            ].astype(out_dtype)
+            data[b, c // out_channels * (inp_channels // groups) + dc, h + dh, w + dw].astype(
+                out_dtype
+            )
             * kernel[
                 c // out_channels * (inp_channels // groups) + dc,
                 c % out_channels,
                 kernel_height - 1 - dh,
-                kernel_width - 1 - dw
+                kernel_width - 1 - dw,
             ].astype(out_dtype),
             axis=[dc, dh, dw],
         ),