benchmark

benchmark copied to clipboard

V2 Performance Signal Detected by TorchBench CI on '1.13.0.dev20220811+cu113'

9

TorchBench CI has detected a performance signal. Base PyTorch version: 1.13.0.dev20220810+cu113 Base PyTorch commit: e1007950484aa1df4a2f87c9c14b514ffd7736a5 Affected PyTorch version: 1.13.0.dev20220811+cu113 Affected PyTorch commit: 3aeb5e4ff9d56ecd680401cfa3f23e97a279efbe Affected Tests: - test_train[hf_BigBird-cpu-eager]: +11.08978% - test_eval[mnasnet1_0-cpu-jit]:...

github-actions[bot]

fix unet script problem

4

when running `python run_sweep.py -m pytorch_unet -t eval -d cuda --jit`, this would raise error `AttributeError: 'RecursiveScriptModule' object has no attribute 'n_classes'` add a final mark to keep this attibute...

zzpmiracle

[correctness check] added use_cosine_similarity as an extra argument

3

The default allclose() correctness check doesn't work for bfloat16 kernels due to precision differences between fp32 and bfloat16. For trt kernels this is addressed by switching to cosine_similarity. Instead of...

snadampal

Performance Signal Detected by TorchBench CI on '1.13.0.dev20220811+cu113'

2

TorchBench CI has detected a performance signal. Base PyTorch version: 1.13.0.dev20220810+cu113 Base PyTorch commit: e1007950484aa1df4a2f87c9c14b514ffd7736a5 Affected PyTorch version: 1.13.0.dev20220811+cu113 Affected PyTorch commit: 3aeb5e4ff9d56ecd680401cfa3f23e97a279efbe Affected Tests: - test_eval[fastNLP-cuda-eager]: -8.62826% - test_train[hf_BigBird-cpu-eager]:...

github-actions[bot]

add TorchExpert

1

Add an argument `--torchexpert` to enable TorchExpert analysis for profiling. Add submodule [TorchExpert](https://github.com/FindHao/TorchExpert). TorchExpert is an analysis tool for profiling results of PyTorch Profiler.

FindHao

AttributeError: 'tuple' object has no attribute 'shape'

1

Command: `python run.py yolov3 -d cpu -t train --bs 1` Error: ```python Running train method from yolov3 on cpu in eager mode. Traceback (most recent call last): File "run.py", line...

Dominic-ZZ

Allred

2

Added BertLarge model that has 16 attention heads, 24 hidden layers and 1024 as hidden size. Added ddp_trainer as an example that allows to benchmark ddp, all_reduce annd non_ddp version...

sanketpurandare

Add world_language_model transformer and lstm

1

Add world_language_model transformer and lstm. They are also used in release testing.

xuzhao9

xuzhao9

xuzhao9