diff --git a/hopper/benchmark_flash_attention_fp8.py b/hopper/benchmark_flash_attention_fp8.py
index 8490d00b3..48fe7b1b2 100644
--- a/hopper/benchmark_flash_attention_fp8.py
+++ b/hopper/benchmark_flash_attention_fp8.py
@@ -44,7 +44,7 @@ def convert_to_cudnn_type(torch_type):
         return cudnn.data_type.INT64
     elif torch_type == torch.float8_e4m3fn:
         return cudnn.data_type.FP8_E4M3
-    elif torch_type == torch.float8_e4m3fn:
+    elif torch_type == torch.float8_e5m2:
         return cudnn.data_type.FP8_E5M2
     else:
         raise ValueError("Unsupported tensor data type.")