search

Linaqruf / kohya-trainer

Adapted from https://note.com/kohya_ss/n/nbf7ce8d80f29 for easier cloning
Apache License 2.0
1.83k stars 300 forks source link

Various fatal errors on XL Colab 'Start Training Cell': ValueError: no metadata & CalledProcessError: Command '['/usr/bin/python3'... returned non-zero exit status 1. #272

Open Plecko opened 1 year ago

Plecko commented 1 year ago

Hi there,

hope you can help! Struggling to work out where i'm going wrong.

I've tried all your XL training colabs, but i get errors with each one. I thought it might be a ram thing, but i have a pro account and on my last test i was using an A100.

I get three errors throughout the process:

First is at 3.4. Bucketing and Latents Caching:

Screenshot 2023-08-07 at 14 37 25

Then 4.4. Start Training

I get two more errors and it stops working.

Screenshot 2023-08-07 at 14 33 58 Screenshot 2023-08-07 at 14 34 04

I've tried all the XL colabs, and each time i get a similar (or the same) error. Feel like i'm following the steps correctly, but obviously not!

Thanks, S

moon47usaco commented 1 year ago

Kindly fix the sdxl version. Same issue as https://github.com/Linaqruf/kohya-trainer/issues/194