logging hardware specific data in userbench (pytorch#2420)

Summary:
currently we are saving the device type in this table ie cuda vs rocm vs cpu etc. We want to further update to allow for hardware specific data ie a100 vs h100. Here we are specifying this for cuda based devices only.

If using cuda we will dynamically find the gpu model via nvidia-smi and log that model.

in this case for H100 we will see "Nvidia H100" and for a100 we will see "NVidia A100"

this will allow us to run the same benchmark on both types of gpu and find the difference in the results quickly via scuba at any given time

Pull Request resolved: pytorch#2420

Reviewed By: danzimm, nmacchioni

Differential Revision: D61229059

Pulled By: adamomainz

fbshipit-source-id: 13ba4e0b8d4f0cf9ae22df54a6c2b6d74e2dcf13

userbenchmark/triton/run.py

@@ -162,8 +162,8 @@ def _run(args: argparse.Namespace, extra_args: List[str]) -> BenchmarkOperatorRe
             else:
                 print(metrics)
         if not hasattr(torch_version, "git_version") and args.log_scuba:
-            from userbenchmark.triton.fb import log_benchmark
-            log_benchmark(metrics)
+            from pytorch.benchmark.fb.run_utils import log_benchmark
+            log_benchmark(metrics, args.op)
         if args.plot:
             try:
                 opbench.plot()