config_class inconsistency for a custom model

[benderama3]

System Info

Default Google colab env
transformers>=4.39.0

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Tested on colab
Works with transformers==4.38.2 but fails with transformers>=4.39.0

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

tokenizer = AutoTokenizer.from_pretrained("ccdv/lsg-bart-base-4096-booksum")
model = AutoModelForSeq2SeqLM.from_pretrained("ccdv/lsg-bart-base-4096-booksum", trust_remote_code=True)

text = "Replace by what you want."*512
pipe = pipeline("text2text-generation", model=model, tokenizer=tokenizer)
generated_text = pipe(
  text, 
  truncation=True, 
  max_length=128, 
  no_repeat_ngram_size=7,
  num_beams=2,
  early_stopping=True
  )

Error message:

ValueError                                Traceback (most recent call last)

[<ipython-input-2-af21733e3710>](https://localhost:8080/#) in <cell line: 4>()
      2 
      3 tokenizer = AutoTokenizer.from_pretrained("ccdv/lsg-bart-base-4096-booksum")
----> 4 model = AutoModelForSeq2SeqLM.from_pretrained("ccdv/lsg-bart-base-4096-booksum", trust_remote_code=True)
      5 
      6 text = "Replace by what you want."*512

1 frames

[/usr/local/lib/python3.10/dist-packages/transformers/models/auto/auto_factory.py](https://localhost:8080/#) in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
    555                 model_class.register_for_auto_class(cls.__name__)
    556             else:
--> 557                 cls.register(config.__class__, model_class, exist_ok=True)
    558             return model_class.from_pretrained(
    559                 pretrained_model_name_or_path, *model_args, config=config, **hub_kwargs, **kwargs

[/usr/local/lib/python3.10/dist-packages/transformers/models/auto/auto_factory.py](https://localhost:8080/#) in register(cls, config_class, model_class, exist_ok)
    581         """
    582         if hasattr(model_class, "config_class") and model_class.config_class != config_class:
--> 583             raise ValueError(
    584                 "The model class you are passing has a `config_class` attribute that is not consistent with the "
    585                 f"config class you passed (model has {model_class.config_class} and you passed {config_class}. Fix "

ValueError: The model class you are passing has a `config_class` attribute that is not consistent with the config class you passed (model has <class 'transformers_modules.ccdv.lsg-bart-base-4096-booksum.e1c052c8cdaf3bd6fc245209e8530c392ea07510.modeling_lsg_bart.LSGBartConfig'> and you passed <class 'transformers_modules.ccdv.lsg-bart-base-4096-booksum.e1c052c8cdaf3bd6fc245209e8530c392ea07510.modeling_lsg_bart.LSGBartConfig'>. Fix one of those so they match!

Both classes are the same.

Expected behavior

No error.

[amyeroberts]

cc @Rocketknight1

[Rocketknight1]

Yeah, this is very likely related to the fix here: #29854

@benderama3 can you please check the PR branch there and tell me if it fixes the issue? We'll try to prioritize merging it if so - we've just been swamped with new model releases!

You can install the PR branch with pip install git+https://github.com/huggingface/transformers.git@update_config_class_check

[Rocketknight1]

@benderama3 this has been merged! You can now get the fix by installing from main with https://github.com/huggingface/transformers.git. Please let me know if this resolves the issue!

[luoxuan-cs]

still not fixed

[Rocketknight1]

@luoxuan-cs Can you give us any code to reproduce the issue you're seeing?