[Question] Getting ZeroDivisionError when launching SimpleTuner

[vedantroy]

Getting the following error:

Traceback (most recent call last):
  File "SimpleTuner/train.py", line 31, in <module>
    trainer.init_data_backend()
  File "SimpleTuner/helpers/training/trainer.py", line 670, in init_data_backend
    self._recalculate_training_steps()
  File "SimpleTuner/helpers/training/trainer.py", line 1090, in _recalculate_training_steps
    self.config.max_train_steps / self.config.num_update_steps_per_epoch
ZeroDivisionError: division by zero

Not sure what's going on. My config has max train steps set to 4500. I'm using 500 images with 500 steps per image, so I'm not sure what logic takes precedence here.

Here are some logs:

2025-01-03 05:43:05,966 [INFO] Moving text encoder to GPU.                                                                                                                                                       
2025-01-03 05:43:06,021 [INFO] Moving text encoder 2 to GPU.                                                                                                                                                     
2025-01-03 05:43:09,325 [INFO] Loading data backend config from /tmp/tmplw7xmf_s/dataset_training_training_1735882968/datasetbackend.json                                                                        
2025-01-03 05:43:09,325 [INFO] Configuring text embed backend: text-embed-cache                                                                                                                                  
Loading pipeline components...: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:00<00:00, 870.51it/s]
2025-01-03 05:43:09,840 [INFO] (Rank: 0) (id=text-embed-cache) Listing all text embed cache entries                                                                                                              
2025-01-03 05:43:09,840 [INFO] Pre-computing null embedding                                                                                                                                                      
2025-01-03 05:43:15,314 [INFO] Completed loading text embed services.                                                                                                                                            
2025-01-03 05:43:15,350 [INFO] Precomputing the negative prompt embed for validations.                                                                                                                           
dict_items([('text-embed-cache', {'config': {}})]) 

On the last line, I printed out StateTracker.get_data_backends().items():

SimpleTuner/helpers/training/trainer.py

Line 1109 in 13c405f

for _, backend in StateTracker.get_data_backends().items()

So, it's clear to me why the error is happening from a stacktrace perspective. But I don't get what's going wrong from a config perspective.

[vedantroy]

This happens if you misconfigure. Specifically no image datasets I think.