diff --git a/fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py b/fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py
index b2b8d0eb7..57d95ad5d 100644
--- a/fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py
+++ b/fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py
@@ -82,7 +82,7 @@ def benchmark(
             # Compute the output given quantized values.
             output = quantize_op.compute(*quantized_vals)
             # Compare the quantize op output to reference as a sanity check.
-            sim_check = torch.mean(torch.pow(torch.abs(output - out_ref), 2))
+            sim_check = torch.mean(torch.pow(output - out_ref, 2))
 
             # Now perform benchmark.
             if bench_quantize: