Migrate Test Infra to pytorch/test-infra
PaliC opened this issue · 1 comments
In the longer run it will be more scalable to use the general pytorch test-infra. Therefore, let's do it when ours is small :)
- Clone the repository
- Modify runtime_nightly.yaml and runtime_tests.yaml to use pytorch/test_infra
- Create a Pull Request and press the button on the PR to import it into Phabricator.
The annoying bit about this issue is that the workflow for building and testing multipy can no longer use docker as running a docker file within a docker file doesn't really work. (pytorch/test_infra creates a docker file with CentOS 7) Therefore, you have to build torch::deploy with the instructions in our README.md and then run the tests we run in the CI. The hardest part about this will be getting the dependencies right as build-requirements.txt is for a debian based system not CentOS 7. Therefore, you'll likely need to figure out the necessary dependencies through trial and error.
There should be with-ssh support which should help with debugging.
I also recommend taking a look at Github actions before starting. But generally with the pytorch/test_infra, you just have to set it up and then run a bash script to build torch::deploy + test it.
If you get stuck anywhere feel free to reach out to Sahan Paliskara.