hiro/ray
1
0
Fork 0
mirror of https://github.com/vale981/ray synced 2025-03-06 10:31:39 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

[Doc] [Serve] Fixed minor typo and removed extract ',' (#22101)

Browse source
This commit is contained in:
Jules S. Damji 2022-02-04 14:51:38 -08:00 • committed by GitHub
parent 5ae8d5b8af
commit c5c5e01b5d
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
doc/source/serve/pipeline.rst
View file

@ -32,13 +32,13 @@ The Serve Pipeline has the following features:
- Each model in the DAG can be scaled to many replicas across the Ray cluster.
You can fine-tune the resource usage to achieve maximum throughput and utilization.
Compare to ServeHandle, Serve Pipeline is more explicit about the dependencies
Compared to ServeHandle, Serve Pipeline is more explicit about the dependencies
of each model in the pipeline and let you deploy the entire DAG at once.
Compare to KServe (formerly KFServing), Serve Pipeline enables writing pipeline
Compared to KServe (formerly KFServing), Serve Pipeline enables writing pipeline
as code and arbitrary control flow operation using Python.
Compare to building your own orchestration micro-services, Serve Pipeline helps
Compared to building your own orchestration micro-services, Serve Pipeline helps
you to be productive in building scalable pipeline in hours.
@ -58,7 +58,7 @@ You can import it as:
You can decorate any function or class using ``pipeline.step``. You can then
combine these steps into a pipeline by calling the decorated steps. In
the example below, we have a single step that takes the special node ``pipeline.INPUT``,
, which is a placeholder for the arguments that will be passed into the pipeline.
which is a placeholder for the arguments that will be passed into the pipeline.
Once you have defined the pipeline by combining one or more steps, you can call ``.deploy()`` to instantiate it.
Once you have instantiated the pipeline, you can call ``.call(inp)`` to invoke synchronously.