missing pad_token_idx in StableLmConfig after 5.0 update

[mario-aws]

System Info

transformers 5.0.0
trainium 2
python 3.10
pytorch 2.9

Who can help?

@ArthurZucker @Cyrilvallez @itazap

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

1. from transformers import StableLmConfig, AutoModelForCausalLM
2. AutoModelForCausalLM.from_config(StableLmConfig())

Error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/shared/mmkrell/vscode/moduscope_ws/TorchNeuronEager/.venv/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 236, in from_config
    return model_class._from_config(config, **kwargs)
  File "/shared/mmkrell/vscode/moduscope_ws/TorchNeuronEager/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 1508, in _from_config
    model = cls(config, **kwargs)
  File "/shared/mmkrell/vscode/moduscope_ws/TorchNeuronEager/.venv/lib/python3.10/site-packages/transformers/models/stablelm/modeling_stablelm.py", line 702, in __init__
    self.model = StableLmModel(config)
  File "/shared/mmkrell/vscode/moduscope_ws/TorchNeuronEager/.venv/lib/python3.10/site-packages/transformers/models/stablelm/modeling_stablelm.py", line 464, in __init__
    self.padding_idx = config.pad_token_id
  File "/shared/mmkrell/vscode/moduscope_ws/TorchNeuronEager/.venv/lib/python3.10/site-packages/transformers/configuration_utils.py", line 164, in __getattribute__
    return super().__getattribute__(key)
AttributeError: 'StableLmConfig' object has no attribute 'pad_token_id'. Did you mean: 'bos_token_id'?

Expected behavior

Normal model creation.

[tobyliu2004]

I can reproduce this and have a fix ready. I've added the missing 'pad_token_id" parameter to 'StableLmConfig' following the same pattern as 'bos_token_id' and 'eos_token_id'. submitting a PR now.

PR: #43573

[mario-aws]

Do we know, why it was not an issue before 5.0?

[tobyliu2004]

After looking at the git history, line 464 self.padding_idx = config.pad_token_id has existed since StableLM was first added in Feb 2024, but pad_token_id was never defined in StableLmConfig. The bug went unnoticed because most users load pretrained models via .from_pretrained() which loads pad_token_id from the saved config.json file. The bug only surfaces when instantiating a fresh config directly with StableLmConfig(). The fix adds the missing attribute following the same pattern as bos_token_id and eos_token_id.

[zucchini-nlp]

cc @Rocketknight1 as you were planning on a PR to cover all models (ref #43539). I prefer to fix all at once than opening separate PR for each model