Scheduler error

1

Got the following error while training my model

---> 95         scheduler.step()
.
.
.
ValueError: Tried to step 1402 times. The specified number of total steps is 1400

But I’ve NOT specified anywhere in my code that the total steps should be 1400

please help

2

You need to provide more information. Maybe share the snippet of your minimal code, the line where the error stack originates, and so on.

However, I remember this sort of error popping up while using pytorch-lightning. There this error occurs when you are using a scheduler in the optimizer without passing the max_epochs argument.

1 Like
3

Here’s the code snipet:

  criterion = nn.BCEWithLogitsLoss()
  optimizer = torch.optim.Adam(model.parameters(),lr=0.001)
  scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer, max_lr=0.001,
                                                steps_per_epoch=int(len(train_dl)),
                                                epochs=num_epochs,
                                                anneal_strategy='linear')
4

You are defining the number of steps via steps_per_epoch and epochs as described in the docs:

  • epochs (int) – The number of epochs to train for. This is used along with steps_per_epoch in order to infer the total number of steps in the cycle if a value for total_steps is not provided. Default: None
  • steps_per_epoch (int) – The number of steps per epoch to train for. This is used along with epochs in order to infer the total number of steps in the cycle if a value for total_steps is not provided. Default: None

and as seen here:

model = nn.Linear(1, 1)

criterion = nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(model.parameters(),lr=0.001)
num_epochs = 10
steps_per_epoch = 2
scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer, max_lr=0.001,
                                                steps_per_epoch=steps_per_epoch,
                                                epochs=num_epochs,
                                                anneal_strategy='linear')

for epoch in range(num_epochs):
    for _ in range(steps_per_epoch):
        optimizer.step()
        scheduler.step()
        
optimizer.step()
scheduler.step()
# ValueError: Tried to step 22 times. The specified number of total steps is 20