Issues with the pretrained model

[qsisi]

Hello!

When I load the pretrained models (after GNN fixed) you offered in #3 , I notice that that value of key 'epoch' doesn't correspond with the value of key 'sheduler':

For KPConv-based:
>>>torch.load('indoor.pth')['epoch']
39
>>>torch.load('indoor.pth')['scheduler']
{'gamma': 0.95, 'base_lrs': [0.005], 'last_epoch': 42, '_step_count': 43, 'verbose': False, '_get_lr_called_within_step': False, '_last_lr': [0.0005799111065000276]}

For Mink_based:
>>>torch.load('sparseIndoor.pth')['epoch']
65
>>>torch.load('sparseIndoor.pth')['scheduler']
{'gamma': 0.99, 'base_lrs': [0.05], 'last_epoch': 75, '_step_count': 76, 'verbose': False, '_get_lr_called_within_step': False, '_last_lr': [0.0235293320792825]}

It is a bit confusing for me that the number of epochs doesn't stay the same in 'epoch' and the information recorded in 'scheduler'. So can you give me some hints about it?

Looking forward to your reply, thank you.