Train stopped at 0%

Models
1

My goal is to train a lora model starting from a database. I used this link to help me with the phyton script: LoRA
So I used the script at this link: diffusers/examples/text_to_image/train_text_to_image_lora.py at main · huggingface/diffusers · GitHub
There are two problems:

  1. the script gives an error if I set num of workers greater than 0
  2. instead if I set a num of workers=0, the script does not give errors but the training remains at 0% even if I change some parameters such as the epoch

I used the command line parameters like in the link, changing lambdalabs/pokemon-blip-captions with lambdalabs/naruto-blip-captions.
What could be the error?

2

image
image1087×159 5.61 KB

This is the error…

3

I discovered that it remains at 0% because the script stops at line 744. I tried to make some prints afterwards and they don’t come out in the logger.
This is the script: diffusers/examples/text_to_image/train_text_to_image_lora.py at e1df77ee1ec0400195ad8dcb2099d137b34c9b9f · huggingface/diffusers · GitHub
Is there a problem initializing the vae variable?

4

Is it possible that the system is running out of memory before it even starts? If you reduce the batch size to one does it run?