AutoTokenizer.from_pretrained crashed：AttributeError: 'dict' object has no attribute 'model_type' (fixed)

[skadiUnderTides]

System Info

Environment

• transformers==4.57.2
• python==3.10

Problem

In tokenization_utils_base.py (around line 2419), _config is loaded from tokenizer_config.json with:

_config = json.load(f)
transformers_version = _config.get("transformers_version")

So _config is a dict.

A few lines below, the logic in 4.57.2 does:

if transformers_version and version.parse(transformers_version) <= version.parse("4.57.2"):
    if _is_local and _config.model_type not in [
        "mistral",
        "mistral3",
        "voxstral",
        "ministral",
        "pixtral",
    ]:
        return tokenizer

Here _config.model_type assumes _config is an object, but it is actually a dict, which causes:

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

Suggested fix

Replace .model_type with a dict-safe access, e.g.:

if transformers_version and version.parse(transformers_version) <= version.parse("4.57.2"):
    if _is_local and _config.get("model_type") not in [
        "mistral",
        "mistral3",
        "voxstral",
        "ministral",
        "pixtral",
    ]:
        return tokenizer

Who can help?

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

Expected behavior

[jrp2014]

I get this too: Blaizzy/mlx-vlm#601 (comment)

The models worked fine with 4.57.1

[vasqu]

Sorry, so many simultaneous issues with essentially the same issue. We added a fix in #42389. It was also added to the patch release here https://github.com/huggingface/transformers/releases/tag/v4.57.3

I'll close this but lmk if this still persists

[jrp2014]

4.57.3 seems to have sorted my various breakages. Thank you.