With the automatic model refresh feature, when the underlying model changes, your pipeline
updates to use the new model. Because the RunInference transform automatically
updates the model handler, you don't need to redeploy the pipeline. With this
feature, you can update your model in real time, even while the Apache Beam
pipeline is running.
Automatic model refresh provides two methods for updating machine learning (ML) models, watch mode and event mode.
Watch mode
Use one of the Apache Beam provided patterns, such as the
WatchFilePattern
class, to watch for the latest file in your Cloud Storage bucket.
WatchFilePattern uses timestamps to match a file_pattern and emits the latest
ModelMetadata,
which the RunInference PTransform uses to update your ML model.
To learn more about using WatchFilePattern to automatically refresh ML
models, see
Use WatchFilePattern to auto-update ML models in RunInference
in the Apache Beam documentation.
Event mode
Connect your pipeline to an unbounded source, such as
Pub/Sub, to send update events directly to the transform, which
initiates a model update. You configure a custom
side input
PCollection that defines the logic for the model update.
To follow a tutorial that demonstrates how to update your model in production by
using a side input PCollection, see
Update ML models in running pipelines.
What's next
Read more about the automatic model refresh feature in the Apache Beam documentation.