ray-project
[Serve] Trim Ray Serve's "Getting Started" Page
Why are these changes needed?
Ray Serve's "Getting Started" page has a lot of content. This change removes the following sections:
- Scaling out
- Fine-grained resource allocation
- Porting FastAPI
Trimmed info from "Scaling out" and "Fine-grained resource allocation" is consolidated into the "Converting to a Ray Serve Deployment" section.
Link to updated "Getting Started" page: https://ray--27670.org.readthedocs.build/en/27670/serve/getting_started.html
Related issue number
Closes #27437.
Checks
- [X] I've signed off every commit(by using the -s flag, i.e.,
git commit -s) in this PR. - [X] I've run
scripts/format.shto lint the changes in this PR. - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/.
- [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
- Testing Strategy
- [X] Unit tests
- Doc tests are updated.
- [X] Unit tests
There are couple if questions that come to the front. Anyone new reading this page quick start might takeaway, from the example demonstrated, that this is all about external pre-trained models and how to use them with or deploy them onto Ray Serve. What if they have questions such as:
- I have an existing scikit-model or xgboost model in pickle format, stored on local FS or cloud, how do I deploy it?
- I have an existing PyTorch or TF model in its native format on disk or cloud, where do I find info to deploy it onto Ray Serve?
- I have experimental models in a model store (MLflow or W & B) and I wish to access them and run them on Ray Serve, how do it?
A reader would have to forage those nuggets either by searching or perusing examples and guides.
Perhaps, have some section toward the end that may link to assets that show them or speak to those concerns.
Anyone new reading this page quick start might takeaway, from the example demonstrated, that this is all about external pre-trained models and how to use them with or deploy them onto Ray Serve.
@dmatrix Good point– I added a snippet in the first section in the documentation saying:
You can also follow along using your own models from any Python framework.
The page also has links to documentation about serving ML models from popular Python frameworks– once at the end of the Model Example: Before Ray Serve section and once at the very end in Next Steps. That should help address concerns that Serve only works for pre-trained models.
@shrekris-anyscale one more lint failure, then LGTM to merge
https://buildkite.com/ray-project/ray-builders-pr/builds/42455#01828a00-5adb-45fc-94f2-7e0775be22c6/9184-9514
