Typo in modeling_utils.py causing checkpoint loading error with Qwen2.5-VL

[tanghme0w]

System Info

Version: 4.52.0.dev0

In modeling_utils.py, line 236, qwen2_5_vl is misspelled as qwem2_5_vl

This will cause the load of _checkpoint_conversion_mapping in line 4113 being skipped and thereby causing param names mismatch during checkpoint load.

Reproduction

simply run the from_pretrained method of Qwen2_5_VLForConditionalGeneration to reproduce.

···
from transformers import Qwen2_5_VLForConditionalGeneration

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
"Qwen/Qwen2.5-VL-32B-Instruct", torch_dtype="auto", device_map="auto"
)
···

Expected behavior

Will report missing parameters and the model will be initialized with random weights.

[2Joey]

lol bro🤣

[zucchini-nlp]

Duplicate of #38056, and will be fixed soon