Update scaled_mm signature in quantize benchmark. (pytorch#2779)

Summary:
Pull Request resolved: pytorch#2779

D58699985 changed the return type of `torch._scaled_mm`. We need to update our implementation to match to avoid errors on main. We should consider adding a test to prevent future API breakages.

Differential Revision: D58981269

fbshipit-source-id: 24891caafbe71bc918f909fae0171b9cfdbda0eb

fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py

@@ -165,7 +165,7 @@ def quantize(self, x, w):
         return xq, wq, x_scale, w_scale
 
     def compute(self, xq, wq, x_scale, w_scale):
-        output, _ = torch._scaled_mm(
+        output = torch._scaled_mm(
             xq,
             wq,
             bias=None,