-
Notifications
You must be signed in to change notification settings - Fork 288
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
How is TorchBench applied to testing new versions of PyTorch? #2499
Comments
You can install nightly pytorch for testing, but the default benchmark only focus on sinlge iteration's computensive part(forward, backward, and optimizer phases). you may check this paper for more details. https://arxiv.org/abs/2304.14226 |
非常感谢您的回复,还想问一下,当pytorch的训练功能有组件更新或者添加了新功能,是如何进行回归测试的? |
If the new features are enabled by default and used in models included in torchbench, the ci will do the testing naturally. Since we use the nightly pytorch, if there are some updates on already applied testings, they will be tested automatically. |
For other new features, the PR authors are usually required to write unit tests in pytorch repo. |
Hello, may I ask what tasks will be used for end-to-end testing before the release of the new version of PyTorch?
Will the test focus on the consistency of metrics between the previous and subsequent versions, such as the loss of training tasks, iteration speed, etc
The text was updated successfully, but these errors were encountered: