V3 Performance Signal Detected by TorchBench Userbenchmark "torch-nightly" on 'b3874abf04e0579fdc0025195593d90b97178f0e'

[github-actions[bot]]

TorchBench CI has detected a performance signal or runtime regression.

Base PyTorch commit: 8c684663042a61de20f163ab6a292cd43c67e294

Affected PyTorch commit: b3874abf04e0579fdc0025195593d90b97178f0e

Affected Tests:

• test_eval[mobilenet_v3_large-cuda-eager]_latency: -13.73389%
• test_eval[yolov3-cuda-eager]_latency: -11.76849%

Tests that were no longer run on affected commit:

Tests that were newly added on affected commit:

Runtime regressions found? No runtime errors were found in the new benchmarks run--you are all good there!

GitHub workflow that triggered this issue: https://github.com/pytorch/benchmark/actions/runs/6027298051

cc @xuzhao9

[xuzhao9]

Bisection: https://github.com/pytorch/benchmark/actions/runs/6042017533

[xuzhao9]

Result:

{
  "target_repo": "pytorch",
  "start": "e68b3ad14f9598334a97c1aff15408e1d8f09002",
  "end": "620d267ef30b199e0ba1b37407dcf8670448fb6a",
  "result": []
}