Machine Learning/LiftWing/API - Wikitech
Jump to content
From Wikitech
Machine Learning
LiftWing
Use the
Lift Wing
API to access Wikimedia machine learning (ML) models. Lift Wing hosts both experimental and production ML models as inference services that you can query for predictions.
Guides
Find and interpret available models
Access Lift Wing API:
Internal clients (inside WMF production network)
External clients (public web, Toolforge, Cloud VPS)
Migrate from ORES to Lift Wing
Reference
API reference
List of current inference services
API features
The Lift Wing API offers the following key features:
Access to predictions from various Wikimedia ML models, via a REST API
Access to multilingual and language-specific models, for a variety of wiki projects
Centrally-managed
MLOps system
and
standardized model serving infrastructure
Lift Wing infrastructure, and the models it hosts, replace
ORES
. To learn more about the differences and how to migrate, see the
ORES to Lift Wing migration guide
Example apps
Suggested edits and microtasks
make it easier and faster to improve Wikipedia. This tool uses multiple ML models to suggest and categorize editing tasks. (
See the code
Generated article descriptions
summarize article topics to help editors write short descriptions. (
See the code
Access policy
You can access Lift Wing model servers through internal or external API endpoints, depending on the model and your access permissions. Use
Machine_Learning/LiftWing#Current_Inference_Services
to see the list of available models. Models in the "experimental" namespace are only available internally.
Internal access
is for clients internal to the production WMF network. It is not available to Toolforge nor Cloud VPS machines.
External access
is available for all clients, including from the public internet, Toolforge, and Cloud VPS.
Rate limits vary for internal and external users; see the links above for details.
Stability policy
See
mw:Wikimedia_APIs/Stability_policy
for general API infrastructure policy.
Models available through the Lift Wing API have varying levels of stability, depending on whether they are experimental or in production. There is no unified policy for Lift Wing services, because each service's stability and SLOs depend on its intended use case. Consult the
model card
to understand the model's existing and intended use cases, along with its freshness. Use the
API reference
to understand the status of specific endpoints.
Report a bug
File a task in Phabricator
using the tag
Machine-Learning-team
See existing tasks on the
team board
Get help
Contact the WMF ML team in one of the following ways:
#wikimedia-ml
on
Libera IRC
Email to
ml@wikimedia.org
Visit
mw:Machine Learning
for more info.
Contribute
See
Machine_Learning/LiftWing#Contributing_to_the_project
Changelog
See
Retrieved from "
Category
Landing page
Machine Learning/LiftWing/API
Add topic