V3 Performance Signal Detected by TorchBench Userbenchmark "torch-nightly" on '7a5da818220cc4c950128db5ea65ec98dece559e'

[github-actions[bot]]

TorchBench CI has detected a performance signal or runtime regression.

Base PyTorch commit: 7726721661ea114acb81a860519d0a1501d88fca

Affected PyTorch commit: 7a5da818220cc4c950128db5ea65ec98dece559e

Affected Tests:

• test_train[phlippe_densenet-cuda-eager]_latency: -10.27443%
• test_eval[phlippe_densenet-cuda-eager]_latency: -9.05751%
• test_train[phlippe_resnet-cuda-eager]_latency: -7.84421%
• test_eval[timm_resnest-cuda-eager]_gmem: +7.04564%
• test_train[torchrec_dlrm-cuda-eager]_latency: -12.48976%

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/5179339828

cc @xuzhao9

[xuzhao9]

2023-06-05

[xuzhao9]

Workflow: https://github.com/pytorch/benchmark/actions/runs/5181829104

[xuzhao9]

{
  "target_repo": "pytorch",
  "start": "39b04370db85313efe23966d04499970a882415c",
  "end": "881307abcfd8fd229fa521098597f4becb2d6c86",
  "result": []
}