Warnings with transformers >=4.46

[AuroraWright]

When run on transformers 4.46 or greater, it shows these warnings:

Config of the encoder: <class 'transformers.models.vit.modeling_vit.ViTModel'> is overwritten by shared encoder config: ViTConfig

Config of the decoder: <class 'transformers.models.bert.modeling_bert.BertLMHeadModel'> is overwritten by shared decoder config: BertConfig

The model still seems to work fine however

[jzhang533]

You can simply ignore this warning. the reason is mangaocr is using different settings, for example, there are 2 hidden layers in mangaocr decoder, but there are 12 hidden layers in the original bert.

if you feel boring, warning messages from transformers could be suppressed by

import logging
loggers = [logging.getLogger(name) for name in logging.root.manager.loggerDict]
for logger in loggers:
    if "transformers" in logger.name.lower():
        logger.setLevel(logging.ERROR)

[MoeMonsuta]

I use use pip install transformers==4.45.2 to avoid long error. Will upgrade when developer resolves issue.

[jzhang533]

I am training the model myself, and this warning disappeared on my trained model, after I changed https://github.com/kha-white/manga-ocr/blob/master/manga_ocr_dev/training/get_model.py to https://github.com/jzhang533/manga-ocr/blob/master/manga_ocr_dev/training/my_get_model.py

I am still trying to train a good model, you can see some examples of my current trained model here: https://github.com/jzhang533/manga-ocr?tab=readme-ov-file#examples

If I could train a descent model, will share to the community.