LR finder returns None

[Borda]

🐛 Bug: LR finder returns None

Could be come model or LR finder range which yield in returning LR as None which crashes the consecutive

To Reproduce

https://www.kaggle.com/jirkaborovec/titanic-crash-with-lightning-flash

from pytorch_lightning.loggers import CSVLogger
from pytorch_lightning.callbacks import StochasticWeightAveraging
from pytorch_lightning import seed_everything

seed_everything(7)
swa = StochasticWeightAveraging(swa_epoch_start=0.6)
logger = CSVLogger(save_dir='logs/')
trainer = flash.Trainer(
    max_epochs=150,
    gpus=torch.cuda.device_count(),
    logger=logger,
    accumulate_grad_batches=12,
    gradient_clip_val=0.01,
    auto_lr_find=True,
)

# ==============================

trainer.tune(model, datamodule=datamodule, lr_find_kwargs=dict(min_lr=1e-4, max_lr=0.1, num_training=65),)
print(f"Learning Rate: {model.learning_rate}")

# ==============================

trainer.fit(model, datamodule=datamodule)

Expected behavior

I see several options:

• raise an error on the LR finder tune so it does not interfere with consecutive fit
• raise just warning and continue with the model's initial LR value
• find why the LR is failing

Additional context

include in Tune redesign #9103

cc @akihironitta @Borda @rohitgr7

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!

[rohitgr7]

not stale

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!

[tchaton]

not stale.

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!

[asicoderOfficial]

It still persists :(

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!