How to implement SWA?

As per the docs, SWA can be

implemented as a callback

swa_callback = StochasticWeightAveraging(swa_lrs=5e-4, swa_epoch_start=1)
trainer = pl.Trainer(callbacks=[swa_callback, ...]

or as a parameter to Trainer
trainer = pl.Trainer(..., stochastic_weight_avg = True, ...)

At the end of the training, I was expecting to models to be saved, a model trained by the trainer and an averaged model generated by SWA. However, I see only one model.

How can I get the averaged model generated by SWA?