Update test

Author: Xiao YU

tests/kernels/attention/test_paged_attention.py

@@ -165,7 +165,7 @@ def run_cuda_benchmark(num_iters: int, profile: bool = False) -> float:
     print("\nresult f16= ", output.shape, output[0,0,:])
     f16_output = output
 
-    diff = f16_output - fp8_ouput
+    diff = f16_output - fp8_output
     print("output mean(fp16, fp8) = ", torch.mean(f16_output), torch.mean(fp8_output))
     print("diff(fp16, fp8) mean, std = ", torch.mean(diff), torch.std(diff))