SDXL LoRA Training Stuck at Step 0

I am encountering an issue where the SDXL LoRA training process stucks step 0, fails to progress, and prematurely increments the epoch count (e.g., current_epoch: 0, epoch: 1). Despite setting num_epochs=40, the training does not complete even a single epoch. The GPU power consumption remains low, indicating underutilization.

Environment

OS: Windows 11

Python: Conda virtual environment (Python 3.10.6)

GPU: [RTX 4070 for laptop]

Dependencies: All packages updated to latest versions (Torch, xformers, etc.)

Script: Custom PowerShell script launching train_network.py with SDXL-specific parameters.


Here's the screenshot,it just stucks here for hours

Start by reproducing the stall with the reported Windows 11, Python 3.10.6, RTX 4070 Laptop, and SDXL parameters in the custom PowerShell script. Inspect train_network.py and the training output around step 0 and epoch counting; done means identifying a reproducible cause and confirming that one epoch progresses normally.

Written by the indexing model from the issue text.