V3 Performance Signal Detected by TorchBench Userbenchmark "torch-nightly" on '7726721661ea114acb81a860519d0a1501d88fca'

[github-actions[bot]]

TorchBench CI has detected a performance signal or runtime regression.

Base PyTorch commit: a14be7981bcef6186441a6c5780976e27e6246ea

Affected PyTorch commit: 7726721661ea114acb81a860519d0a1501d88fca

Affected Tests:

• test_train[detectron2_maskrcnn-cuda-eager]_latency: +9.25119%
• test_eval[doctr_det_predictor-cuda-eager]_cmem: +9.67283%
• test_train[lennard_jones-cuda-eager]_latency: +9.45848%
• test_train[mnasnet1_0-cuda-eager]_latency: +8.21508%
• test_train[mobilenet_v3_large-cuda-eager]_latency: +8.42297%
• test_eval[opacus_cifar10-cuda-eager]_latency: +8.53870%
• test_train[phlippe_densenet-cuda-eager]_latency: +14.65026%
• test_eval[phlippe_densenet-cuda-eager]_latency: +7.71325%
• test_train[phlippe_resnet-cuda-eager]_latency: +12.35126%
• test_eval[pyhpc_turbulent_kinetic_energy-cuda-eager]_latency: +9.91995%
• test_eval[timm_resnest-cuda-eager]_gmem: -13.08281%
• test_train[torchrec_dlrm-cuda-eager]_latency: +17.33417%
• test_eval[torchrec_dlrm-cuda-eager]_latency: +7.35767%
• test_train[tts_angular-cuda-eager]_latency: +9.57081%

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/5164520717

cc @xuzhao9

[xuzhao9]

2023-06-03

Bisection workflow: https://github.com/pytorch/benchmark/actions/runs/5176866505

Bisection workflow 2: https://github.com/pytorch/benchmark/actions/runs/5179562964

Bisection workflow 3: https://github.com/pytorch/benchmark/actions/runs/5181786643

[xuzhao9]

Result:

{
  "target_repo": "pytorch",
  "start": "bebb8b7c1e6f3baf35c00d2a9265a285b4e439a8",
  "end": "39b04370db85313efe23966d04499970a882415c",
  "result": []
}