fix: Use model name for AutoProcessor to fix formula enrichment AttributeError

[b-g-d]

Fixes #2681

Problem

When do_formula_enrichment=True is set, Docling fails with:

AttributeError: 'dict' object has no attribute 'model_type'

Root Cause

Passing a local Path object to AutoProcessor.from_pretrained() causes
transformers to load the tokenizer config as a dict, but then tries to access
_config.model_type as an object attribute.

Solution

Use the model name ('docling-project/CodeFormulaV2') instead of the local
path. Transformers will automatically use the cached model, and the config
will be properly loaded as an object.

Testing

• Tested with do_formula_enrichment=True - no longer throws AttributeError
• Verified formulas are still extracted correctly
• Confirmed cached model is still used (no re-download)

[github-actions]

✅ DCO Check Passed

Thanks @b-g-d, all your commits are properly signed off. 🎉

[dosubot]

Related Documentation

Checked 4 published document(s) in 1 knowledge base(s). No updates required.

^{How did I do? Any feedback?}  

[mergify]

Merge Protections

Your pull request matches the following merge protections and will not be merged until they are valid.

🟢 Enforce conventional commit

Wonderful, this rule succeeded.

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[b-g-d]

Update: The underlying transformers bug has been fixed in transformers 4.57.3 (released 2025-11-25). However, this PR still provides value by:

1. Using explicit model names - More maintainable and clear than local paths
2. Better code clarity - Makes it obvious which model is being loaded
3. Works with all transformers versions - Not dependent on the transformers fix

The fix in transformers changed _config.model_type to _config.get("model_type") to properly handle dict configs when loading from local paths.

This PR can still be merged as it represents a best practice, even though the immediate bug is resolved in transformers 4.57.3+.

[dolfim-ibm]

@b-g-d The problem with your proposed solution is that Docling should be able to load the models from an artifacts directory which is not the HF cache dir.