fix: restore missing tokenizer_class attribute in ModelInfos.__init__

[Ker102]

Summary

Fixes #44661 — transformers add-new-model-like crashes with AttributeError: 'ModelInfos' object has no attribute 'tokenizer_class' when selecting a model that is in TOKENIZER_MAPPING_NAMES.

Root Cause

PR #40936 refactored TOKENIZER_MAPPING_NAMES from (slow_class, fast_class) tuples to single fast tokenizer class names. During this change, self.tokenizer_class was removed from the if branch in ModelInfos.__init__ (line 144), but get_user_input() still accesses it at line 719:

if old_model_infos.tokenizer_class is not None:

The else branch (when the model is NOT in TOKENIZER_MAPPING_NAMES) correctly sets self.tokenizer_class = None, but the if branch (when it IS in the mapping) never assigns it at all.

Fix

Add self.tokenizer_class = None in the if branch of ModelInfos.__init__, matching the behavior of the else branch:

  if self.lowercase_name in TOKENIZER_MAPPING_NAMES:
+     self.tokenizer_class = None
      self.fast_tokenizer_class = TOKENIZER_MAPPING_NAMES[self.lowercase_name]

Reproduction

transformers add-new-model-like
# Select qwen2_5_vl (or any model in TOKENIZER_MAPPING_NAMES)
# → AttributeError: 'ModelInfos' object has no attribute 'tokenizer_class'

[Rocketknight1]

No drive-by code agent PRs! https://github.com/huggingface/transformers/blob/main/AGENTS.md#coordination-before-coding

[Ker102]

Sorry, I had only read the CONTRIBUTING.md file, I just wanted to provide a solution to this issue, could I still work on it?