lightning/docs/Trainer/Testing loop.md

To ensure you don't accidentally use test data to guide training decisions Lightning makes running the test set deliberate.

---

test

You have two options to run the test set. First case is where you test right after a full training routine.

# run full training
trainer.fit(model)

# run test set
trainer.test()

Second case is where you load a model and run the test set

model = MyLightningModule.load_from_metrics(
    weights_path='/path/to/pytorch_checkpoint.ckpt',
    tags_csv='/path/to/test_tube/experiment/version/meta_tags.csv',
    on_gpu=True,
    map_location=None
)

# init trainer with whatever options
trainer = Trainer(...)
    
# test (pass in the model)
trainer.test(model)

In this second case, the options you pass to trainer will be used when running the test set (ie: 16-bit, dp, ddp, etc...)