V3 Performance Signal Detected by TorchBench Userbenchmark "torch-nightly" on 'd902c353e928c8dccc1833b999fb97dfdde7ab12'

[github-actions[bot]]

TorchBench CI has detected a performance signal or runtime regression.

Base PyTorch commit: 6c94bfc7da1b91f110929dfb7f8b5fa6d6baaa0a

Affected PyTorch commit: d902c353e928c8dccc1833b999fb97dfdde7ab12

Affected Tests:

• test_eval[detectron2_fasterrcnn_r_50_dc5-cuda-eager]_latency: -20.32668%
• test_train[pytorch_CycleGAN_and_pix2pix-cuda-eager]_latency: -10.66427%
• test_eval[shufflenet_v2_x1_0-cuda-eager]_latency: -10.29876%
• test_eval[yolov3-cuda-eager]_latency: -14.08622%

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/6549849555

cc @xuzhao9

[xuzhao9]

Bisection: https://github.com/pytorch/benchmark/actions/runs/6553686295

[xuzhao9]

{
  "target_repo": "pytorch",
  "start": "4a388e70f2c7fedcdef598fb2144cc593452f04d",
  "end": "013b51f8cc6f973d85c5adb36d5c110b7d42c7a2",
  "result": []
}

Closed.