Fix None quantization_config equivalence with omitted param in AutoModel.from_pretrained

[albertvillanova]

Fix quantization_config=None equivalence with omitted parameter in AutoModel.from_pretrained.

Currently, AutoModel.from_pretrained treats differently missing and None quantization_config. I think passing None should be treated the same as omitting the key altogether, not as an invalid quantization config.

This PR:

• Fixes inconsistent handling of quantization_config=None vs omitted parameter in AutoModel.from_pretrained
• Ensures quantization_config=None behaves identically to omitting the parameter entirely

Problem

When using TRL CLI with GPT OSS models, calls like:

  model = AutoModelForCausalLM.from_pretrained("my_model", quantization_config=None)

would fail with AttributeError: 'NoneType' object has no attribute 'to_dict', differently than:

model = AutoModelForCausalLM.from_pretrained("my_model")

Error traceback:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/fsx/albert/dev/transformers/src/transformers/models/auto/auto_factory.py", line 549, in from_pretrained
    config, kwargs = AutoConfig.from_pretrained(
  File "/fsx/albert/dev/transformers/src/transformers/models/auto/configuration_auto.py", line 1323, in from_pretrained
    return config_class.from_dict(config_dict, **unused_kwargs)
  File "/fsx/albert/dev/transformers/src/transformers/configuration_utils.py", line 839, in from_dict
    logger.info(f"Model config {config}")
  File "/fsx/albert/dev/transformers/src/transformers/configuration_utils.py", line 873, in __repr__
    return f"{self.__class__.__name__} {self.to_json_string()}"
  File "/fsx/albert/dev/transformers/src/transformers/configuration_utils.py", line 985, in to_json_string
    config_dict = self.to_diff_dict()
  File "/fsx/albert/dev/transformers/src/transformers/configuration_utils.py", line 887, in to_diff_dict
    config_dict = self.to_dict()
  File "/fsx/albert/dev/transformers/src/transformers/configuration_utils.py", line 964, in to_dict
    self.quantization_config.to_dict()
AttributeError: 'NoneType' object has no attribute 'to_dict'

See comment in downstream hotfix:

• 🗜 Hotfix: avoid passing quantization_config=None trl#4019 (review)

CC: @qgallouedec

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: auto

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.