From b4c3f7052955c59f05c6731000fd036cee89fc3a Mon Sep 17 00:00:00 2001 From: Yukio Siraichi Date: Thu, 26 Sep 2024 15:59:18 -0300 Subject: [PATCH] Add nigthly CI section. --- docs/torchbench.md | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/docs/torchbench.md b/docs/torchbench.md index 3d29fc57877..4895497381d 100644 --- a/docs/torchbench.md +++ b/docs/torchbench.md @@ -505,6 +505,15 @@ open an issue. Our benchmarking scripts allow for the specification of XLA (actual compiler) flags by passing `--xla-flags=`. By default, we run without any specific flags. +## Nightly CI + +Every night, we run Torchbench as described in this document in our Nightly +CI. Specifically, we run the following configurations for inference and training, using +the default for the options not mentioned: + +- on TPU: dynamo and non-dynamo +- on GPU: eager, inductor, dynamo, and non-dynamo + [1]: https://github.com/pytorch/benchmark [2]: https://github.com/pytorch/pytorch/blob/main/benchmarks/dynamo/torchbench.py