Training models directly on your datastore
Model instances may be trained if a trainer is set on the Model when db.apply is called.
When models are trained, if CFG.cluster.compute has been configured with a ray scheduler, then superduper deploys a job on the connected ray cluster.
Basic pattern
from superduper.ext.<framework> import <Framework>Trainer
from superduper.ext.<framework> import <ModelCls>
db.apply(
<ModelCls>(
*args,
trainer=<Framework>Trainer(**trainer_kwargs),
**kwargs,
)
)
Fitting/ training models by framework
Not all Model types are trainable. We support training for the following frameworks:
| Framework | Training Link |
|---|---|
| Scikit-Learn | link |
| PyTorch | link |
| Transformers | link |