num_training_batches is inf in configure_optimizers #16060
Description
Bug description
The value of num_training_batches is inf when referenced in configure_optimizers(). It seems that it doesn't actually get its correct value until some point later. This causes a very hard-to-find issue because the training runs without error, except the loss is nan.
Something inside optim.lr_scheduler.CyclicLR actually sets the lr of the optimizer to nan.
It would be nice if:
- This value was available
configure_optimizers()was called, or - There was a warning if accessing it before it's set
How to reproduce the bug
import os
import torch
from torch.utils.data import DataLoader, Dataset
from pytorch_lightning import LightningModule, Trainer
class RandomDataset(Dataset):
def __init__(self, size, length):
self.len = length
self.data = torch.randn(length, size)
def __getitem__(self, index):
return self.data[index]
def __len__(self):
return self.len
class BoringModel(LightningModule):
def __init__(self):
super().__init__()
self.layer = torch.nn.Linear(32, 2)
def forward(self, x):
return self.layer(x)
def training_step(self, batch, batch_idx):
loss = self(batch).sum()
self.log("train_loss", loss)
return {"loss": loss}
def validation_step(self, batch, batch_idx):
loss = self(batch).sum()
self.log("valid_loss", loss)
def test_step(self, batch, batch_idx):
loss = self(batch).sum()
self.log("test_loss", loss)
def configure_optimizers(self):
optimizer = torch.optim.SGD(self.layer.parameters(), lr=0.1)
print(f"{optimizer.param_groups[0]['lr'] = }") # 0.1
lr_scheduler = torch.optim.lr_scheduler.CyclicLR(
optimizer=optimizer,
base_lr=0.01,
max_lr=0.1,
step_size_up=self.trainer.num_training_batches * 1, # problematic!
step_size_down=self.trainer.num_training_batches * 2, # problematic!
cycle_momentum=False,
)
print(f"{optimizer.param_groups[0]['lr'] = }") # nan
return [optimizer], [lr_scheduler]
def run():
train_data = DataLoader(RandomDataset(32, 64), batch_size=2)
val_data = DataLoader(RandomDataset(32, 64), batch_size=2)
test_data = DataLoader(RandomDataset(32, 64), batch_size=2)
model = BoringModel()
trainer = Trainer(
default_root_dir=os.getcwd(),
limit_train_batches=1,
limit_val_batches=1,
limit_test_batches=1,
num_sanity_val_steps=0,
max_epochs=1,
enable_model_summary=False,
enable_checkpointing=False,
)
trainer.fit(model, train_dataloaders=train_data, val_dataloaders=val_data)
trainer.test(model, dataloaders=test_data)
if __name__ == "__main__":
run()Error messages and logs
The main hint something is wrong is actually tensorboard printing "NaN or Inf found in input tensor" - but even that doesn't come with a trace telling me who's printing this.
Environment
Current environment
#- Lightning Component (e.g. Trainer, LightningModule, LightningApp, LightningWork, LightningFlow):
#- PyTorch Lightning Version (e.g., 1.5.0):
#- Lightning App Version (e.g., 0.5.2):
#- PyTorch Version (e.g., 1.10):
#- Python version (e.g., 3.9):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):
#- Running environment of LightningApp (e.g. local, cloud):
More info
No response