kohya_ss icon indicating copy to clipboard operation
kohya_ss copied to clipboard

Configs that specify custom `pretrained_model_name_or_path` saved before #292 fail to load

Open evanjs opened this issue 1 year ago • 0 comments

If a user attempts to load a config saved before #292 (?) and they have a custom model selected for Pretrained model name or path, Gradio chokes upon trying to load the config, and endlessly prints the following error.

Error from Gradio
Traceback (most recent call last):
  File "...\kohya_ss\venv\lib\site-packages\gradio\routes.py", line 384, in run_predict
    output = await app.get_blocks().process_api(
  File "...\kohya_ss\venv\lib\site-packages\gradio\blocks.py", line 1027, in process_api
    data = self.postprocess_data(fn_index, result["prediction"], state)
  File "...\kohya_ss\venv\lib\site-packages\gradio\blocks.py", line 939, in postprocess_data
    if predictions[i] is components._Keywords.FINISHED_ITERATING:
TypeError: 'NoneType' object is not subscriptable

It seems like we don't handle older configs properly, and the model's value [] is passed as is. Manually updating old configs to specify "custom" instead of [] seems to work fine.

- "model_list": [],
+ "model_list": "custom",

This seemed to be the only change I needed to make to get things working again. The lack of a value for LoRA_type did not seem to cause any issues.

evanjs avatar Mar 04 '23 16:03 evanjs