diff --git a/tests/kernels/attention/test_cache.py b/tests/kernels/attention/test_cache.py
index a130f9acb..4ff1e590a 100644
--- a/tests/kernels/attention/test_cache.py
+++ b/tests/kernels/attention/test_cache.py
@@ -242,7 +242,7 @@ def test_reshape_and_cache_flash(
     value_cache_compact = permute_and_compact(value_cache)
 
     def convert_fp8_local(output, input, scale, kv_dtype):
-        fp8_input = input.view(torch.float8_e4m3fn)
+        fp8_input = input.view(current_platform.fp8_dtype())
         if scale.numel() == 1:  # per-tensor
             result = scaled_dequantize(
                 fp8_input.flatten(0, 2), scale, group_shape=None, out_dtype=output.dtype