V3 Performance Signal Detected by TorchBench Userbenchmark "torch-nightly" on 'c1406a99df2df9c06e8c7029e2eac41d5b2240cf'

[github-actions[bot]]

TorchBench CI has detected a performance signal or runtime regression.

Base PyTorch commit: 4d648e450b8e1386c0079f22c38aebc14fb93872

Affected PyTorch commit: c1406a99df2df9c06e8c7029e2eac41d5b2240cf

Affected Tests:

• test_train[phlippe_densenet-cuda-eager]_latency: +7.46135%
• test_eval[pytorch_CycleGAN_and_pix2pix-cuda-eager]_latency: +18.45467%
• test_eval[vision_maskrcnn-cuda-eager]_latency: -73.90283%

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/5213333440

cc @xuzhao9

[xuzhao9]

{ "target_repo": "pytorch", "start": "872fdb329b75c832c2cd1314b3613e28289d8fac", "end": "d5142c52d3592e4a5777834be80b276ed1073012", "result": [] }