processing_qwen2_5vl_casa.py

from typing import Any

from transformers.models.qwen2.tokenization_qwen2 import Qwen2Tokenizer

from .processing import BaseProcessor, QwenImageProcessor


class QwenCASAProcessor(BaseProcessor):
    attributes = ["tokenizer"]
    tokenizer_class = "Qwen2Tokenizer"

    def __init__(
        self,
        tokenizer: Qwen2Tokenizer,
        pre_image_tokens: tuple[int, ...] = (151652,),
        post_image_tokens: tuple[int, ...] = (151653,),
        system_start_tokens: tuple[int, ...] = (151644, 8948, 198),
        system_end_tokens: tuple[int, ...] = (151645, 198),
        user_start_tokens: tuple[int, ...] = (151644, 872, 198),
        user_end_tokens: tuple[int, ...] = (151645, 198),
        asst_start_tokens: tuple[int, ...] = (151644, 77091, 198),
        asst_end_tokens: tuple[int, ...] = (151645, 198),
        image_size: int = 448,
        **kwargs: Any,
    ):
        del kwargs
        super().__init__(
            tokenizer=tokenizer,
            pre_image_tokens=pre_image_tokens,
            post_image_tokens=post_image_tokens,
            system_start_tokens=system_start_tokens,
            system_end_tokens=system_end_tokens,
            user_start_tokens=user_start_tokens,
            user_end_tokens=user_end_tokens,
            asst_start_tokens=asst_start_tokens,
            asst_end_tokens=asst_end_tokens,
        )

        self._image_processor = QwenImageProcessor(img_size=image_size)