diff --git a/mmcv/ops/fused_bias_leakyrelu.py b/mmcv/ops/fused_bias_leakyrelu.py
index fe17d2db7..5c0e6d92d 100644
--- a/mmcv/ops/fused_bias_leakyrelu.py
+++ b/mmcv/ops/fused_bias_leakyrelu.py
@@ -257,7 +257,7 @@ def fused_bias_leakyrelu(input: torch.Tensor,
     Returns:
         torch.Tensor: Feature map after non-linear activation.
     """
-
+    assert scale > 0, 'The scale should be greater than 0'
     if not input.is_cuda and input.device.type != 'npu':
         return bias_leakyrelu_ref(input, bias, negative_slope, scale)