Python环境下能否以类的形式调用，并在__init__的时候把模型加载显存（内存）中

[ShiShuyang]

我修改了官网的标程序，Miner能够如预期地运行：

import os
from mineru.cli.common import convert_pdf_bytes_to_bytes_by_pypdfium2, read_fn
from mineru.data.data_reader_writer import FileBasedDataWriter
from mineru.utils.enum_class import MakeMode
from mineru.backend.pipeline.pipeline_analyze import doc_analyze as pipeline_doc_analyze
from mineru.backend.pipeline.pipeline_middle_json_mkcontent import union_make as pipeline_union_make
from mineru.backend.pipeline.model_json_to_middle_json import result_to_middle_json as pipeline_result_to_middle_json
def do_parse(
    pdfpath: bytes,  # List of PDF bytes to be parsed
    p_lang: str='ch',  # List of languages for each PDF, default is 'ch' (Chinese)
    parse_method='auto',  # The method for parsing PDF, default is 'auto'
    p_formula_enable=False,  # Enable formula parsing
    p_table_enable=False,  # Enable table parsing
) -> list[dict]:

    os.environ['MINERU_MODEL_SOURCE'] = 'modelscope'
    os.environ['MINERU_DEVICE_MODE'] = 'cpu'

    #pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(read_fn(pdfpath), 0, None)
    pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(read_fn(pdfpath), 0, 6)
    
    infer_results, all_image_lists, all_pdf_docs, lang_list, ocr_enabled_list = pipeline_doc_analyze([pdf_bytes], [p_lang], parse_method=parse_method, formula_enable=p_formula_enable,table_enable=p_table_enable)
    image_writer = FileBasedDataWriter()
    images_list = all_image_lists[0]
    pdf_doc = all_pdf_docs[0]
    _lang = lang_list[0]
    _ocr_enable = ocr_enabled_list[0]
    middle_json = pipeline_result_to_middle_json(infer_results[0], images_list, pdf_doc, image_writer, _lang, _ocr_enable, p_formula_enable)

    pdf_info = middle_json["pdf_info"]
    content_list = pipeline_union_make(pdf_info, MakeMode.CONTENT_LIST)
    return content_list

但是每次运行似乎都要重新载入资源。特别在logging输出以下内容的时候特别慢：

Downloading Model from https://www.modelscope.cn to directory: C:\Users\Admin\.cache\modelscope\hub\models\OpenDataLab\PDF-Extract-Kit-1.0
2025-07-04 13:28:28,809 - modelscope - INFO - Creating symbolic link [C:\Users\Admin\.cache\modelscope\hub\models\OpenDataLab\PDF-Extract-Kit-1.0].
2025-07-04 13:28:28,809 - modelscope - WARNING - Failed to create symbolic link C:\Users\Admin\.cache\modelscope\hub\models\OpenDataLab\PDF-Extract-Kit-1.0 for C:\Users\Admin\.cache\modelscope\hub\models\OpenDataLab\PDF-Extract-Kit-1___0.

我已经下载好了模型，但是似乎每次还要联网验证？是否可以force_local？以及我不清楚这个Creating symbolic link是否大幅拖慢了速度。尽管Creating symbolic link并没有成功，但是输出的对的。是否可以不symbolic link。

我希望的是，可以用类的形式调用。在实例化的时候就把所有资源在内存显存中存好不卸载，然后调用。比如

Class MinerUAPI:
  def __init__(self):
    self.load_torch = load_some_model
  def run(self, filename: str):
    pdf_bytes = convert_pdf_bytes_to_bytes_by_pypdfium2(read_fn(pdfpath), 0, 6)
    return self.load_torch(pdf_bytes)

[myhloli]

python api的方式是不会重复加载模型的，你可以试下循环调用方法

[ShiShuyang]

大大大大大大佬！
居然是真的，可能是我之前监视内存显存的时候看错了。
非常感谢回复♪(･ω･)ﾉ

[ShiShuyang]

我看错了。关了。