Linaqruf / kohya-trainer

Adapted from https://note.com/kohya_ss/n/nbf7ce8d80f29 for easier cloning
Apache License 2.0
1.84k stars 302 forks source link

The scirpt you used in Section 4.2 is fed with wrong args. #215

Closed zschaoihen closed 1 year ago

zschaoihen commented 1 year ago

When you invoke the "Waifu Diffusion 1.4 Tagger V2", the command you're current using is this:

python tag_images_by_wd14_tagger.py \ "{train_data_dir}" \ --batch_size {batch_size} \ --repo_id {model} \ --thresh {threshold} \ --caption_extension .txt \ --max_data_loader_n_workers {max_data_loader_n_workers}

But the thresh args will not work since it only takes either a general threshold or a character threshold. image

Linaqruf commented 1 year ago

which notebook?

{threshold} are from old notebook

zschaoihen commented 1 year ago

Oh, I must use the old version of LoRA_Dreambooth, but I think it will automatically pull the latest git repo, that's why I cannot make section 4.2 work.