[Question]: uie-x 循环推理显存增长问题

[Yian320]

请提出你的问题

使用paddlenlp uiex模型进行图片信息抽取，将一批图片循环批量进行预测抽取过程中，出现显存增加问题怎么解决，同一批图片单独进行OCR通用识别时不会出现显存增长问题

[zhh8689]

请问你有解决吗？我也遇到了一样的问题

[Yian320]

请问你有解决吗？我也遇到了一样的问题

你好，未解决，你有解决办法吗

[github-actions[bot]]

This issue is stale because it has been open for 60 days with no activity. 当前issue 60天内无活动，被标记为stale。

[w5688414]

请问能提供一下最小复现代码吗？另外推理部署推荐使用fastdeploy https://github.com/PaddlePaddle/FastDeploy/blob/develop/examples/text/uie/README_CN.md?plain=1

[Yian320]

class UIE_X(STD): def init(self, logger=None, config=None): super().init(logger=logger, config=config) if self.config.use_gpu is False: self.device = 'cpu' else: self.device = 'cuda' os.environ["CUDA_VISIBLE_DEVICES"] = config.gpu_id model_uie_x = self.config.model_uiex self.image_info_extractor = Taskflow("information_extraction", task_path=model_uie_x, device_id=int(self.config.gpu_id))

def uie_x_predict(self, image, certificate_type):
    """
    证书图片抽取模型推理
    :param image: 图片
    :param certificate_type: 图片类型
    :return:
    """
    try:
        if str(certificate_type).isdigit():
            schema_num = certificate_type
        else:
            schema_num = self.config.certifi_zh_map.get(certificate_type, "")
        schema_str = f"schema_{schema_num}"
        schema = self.config.schema.get(schema_str, [])
        # print(schema)
        self.image_info_extractor.set_schema(schema)
        # print(image)
        result = self.image_info_extractor({"doc": image})
        # print("result-uie-x", result)
        self.logger.info("图片信息抽取完成，正在返回结果...")
        data = self.extract_data_uie_x_predict_result(schema, result)
        return data
    except Exception as e:
        self.logger.error(f'error file:{e.__traceback__.tb_frame.f_globals["__file__"]} line:{e.__traceback__.tb_lineno} {e}')
        print(f'error file:{e.__traceback__.tb_frame.f_globals["__file__"]} line:{e.__traceback__.tb_lineno} {e}')
        return {}

@w5688414 使用ocr V4版本，看到有说是应为paddleocr 是显存贪婪机制，上述推理代码在处理多组图片时会出现显存增长，目前使用flask部署推理的，测试中显存最大增长到17G

[github-actions[bot]]

This issue is stale because it has been open for 60 days with no activity. 当前issue 60天内无活动，被标记为stale。

[github-actions[bot]]

This issue was closed because it has been inactive for 14 days since being marked as stale. 当前issue 被标记为stale已有14天，即将关闭。