Fix a metrics reporting issue for FP8 row-wise

Summary: It seems that ` metrics.latency` and ` metrics.tflops` are not lists but a scalar.

Reviewed By: xuzhao9

Differential Revision: D58599542

fbshipit-source-id: d4dbfb01f20e44355b01f0f6bebe7dca86c3615e

torchbenchmark/operators/fp8_gemm_rowwise/operator.py

@@ -120,7 +120,7 @@ def tflops(
  m, k = xq.size()
  n, k = wq.size()
  flops = m * k * 2 * n
- return [flops / x / 1e12 * 1e3 for x in metrics.latency]
+ return flops / metrics.latency / 1e12 * 1e3
 
  @register_x_val(label="(M, N, K)")
  def get_x_val(self, example_inputs) -> Tuple[int, int, int]: