processing_qwen3vl_visual.py

from transformers import AutoTokenizer
from transformers.models.qwen3_vl.processing_qwen3_vl import Qwen3VLProcessor


class Qwen3VLVisualProcessor(Qwen3VLProcessor):

    @classmethod
    def _get_arguments_from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
        args = super()._get_arguments_from_pretrained(pretrained_model_name_or_path, **kwargs)
        tok_kwargs = dict(kwargs)
        tok_kwargs.setdefault("trust_remote_code", True)
        visual_tokenizer = AutoTokenizer.from_pretrained(
            pretrained_model_name_or_path, **tok_kwargs
        )
        for i, attr_name in enumerate(cls.attributes):
            if attr_name == "tokenizer":
                args[i] = visual_tokenizer
                break
        return args