dsProject/dsLightRag/Util/GGB/Backup/GGB_3_QVQ_Test.py

import logging

from Util.ObsUtil import ObsUploader

prompt = """
### 几何图形识别专家指令（输入：纯几何图形照片）

**任务目标**
精确提取图形中的几何元素及其空间关系，为GeoGebra重建建立数学模型

## 一、坐标系建立规则（必须遵守）
1. 原点设定：
   - 若存在明显顶点，选最左下角的点为原点(0,0)
   - 若图形对称，选对称中心为原点
   - 示例：原点O = 三角形ABC的顶点A

2. 坐标轴定向：
   - 优先顺序：水平线段 > 垂直线段 > 最长线段
   - 具体规则：
     if 存在水平线段: 以该线段为x轴正方向
     elif 存在垂直结构: 以最左侧垂直线为y轴
     else: 以最长线段为基准轴

## 二、元素列举
1. 按点，线，三角形，四边形，梯形，平行四边形，矩形，正方形，圆等由简单到复杂的顺序列举所有图形
2. 详细描述元素之间的关系，比如点D在线段AB上
3. 详细描述元素之间的位置关系，比如D 在A点正上方，B在CD边的上方中间位置

"""
from openai import OpenAI
from Config.Config import ALY_LLM_BASE_URL, ALY_LLM_API_KEY, OBS_AK, OBS_SERVER, OBS_SK, OBS_PREFIX, OBS_BUCKET

logger = logging.getLogger(__name__)


# 批量处理图片
def batch_qvq(output_dir, img_list):
    img_url_list = []
    for file_path in img_list:
        # 创建上传器实例
        uploader = ObsUploader(OBS_AK, OBS_SK, "https://" + OBS_SERVER)
        # 上传参数
        object_key = OBS_PREFIX + "/" + file_path
        # 执行上传
        success, result = uploader.upload_file(OBS_BUCKET, object_key, file_path)
        # 处理结果
        if success:
            logger.info(f'{file_path}上传成功!')
            # 获取上传文件的 URL
            file_url = f"https://{OBS_BUCKET}.{OBS_SERVER}/{object_key}"
            img_url_list.append(file_url)
        else:
            logger.error(f'{file_path}上传失败!')
            if 'errorCode' in result:
                logger.info(f'错误代码: {result["errorCode"]}')
                logger.info(f'错误信息: {result["errorMessage"]}')
            else:
                logger.error(f'错误信息: {result["error"]}')
    # 多张图片开始解析
    answer_content = ""
    for img_url in img_url_list:
        answer_content = answer_content + qvq_single(img_url)

    # 保存结果到JSON文件
    qvq_result = f"{output_dir}/QvqResult.json"
    with open(qvq_result, "w", encoding='utf-8') as f:
        f.write(answer_content)
    return qvq_result


def qvq_single(image_url):
    # 初始化OpenAI客户端
    client = OpenAI(
        api_key=ALY_LLM_API_KEY,
        base_url=ALY_LLM_BASE_URL
    )
    reasoning_content = ""  # 定义完整思考过程
    answer_content = ""  # 定义完整回复
    is_answering = False  # 判断是否结束思考过程并开始回复

    # 创建聊天完成请求
    completion = client.chat.completions.create(
        model="qvq-max",  # 此处以 qvq-max 为例，可按需更换模型名称
        messages=[
            {
                "role": "user",
                "content": [
                    {
                        "type": "image_url",
                        "image_url": {
                            "url": image_url
                        },
                    },
                    {"type": "text",
                     "text": prompt},
                ],
            },
        ],
        stream=True,
    )

    print("\n" + "=" * 20 + "思考过程" + "=" * 20 + "\n")

    for chunk in completion:
        # 如果chunk.choices为空，则打印usage
        if not chunk.choices:
            print("\nUsage:")
            print(chunk.usage)
        else:
            delta = chunk.choices[0].delta
            # 打印思考过程
            if hasattr(delta, 'reasoning_content') and delta.reasoning_content != None:
                print(delta.reasoning_content, end='', flush=True)
                reasoning_content += delta.reasoning_content
            else:
                # 开始回复
                if delta.content != "" and is_answering is False:
                    print("\n" + "=" * 20 + "完整回复" + "=" * 20 + "\n")
                    is_answering = True
                # 打印回复过程
                print(delta.content, end='', flush=True)
                answer_content += delta.content

    # print("=" * 20 + "完整思考过程" + "=" * 20 + "\n")
    # print(reasoning_content)
    # print("=" * 20 + "完整回复" + "=" * 20 + "\n")
    answer_content = answer_content.replace("```json", "")
    answer_content = answer_content.replace("```", "")
    return answer_content


if __name__ == '__main__':
    img_url = "https://dsideal.obs.cn-north-1.myhuaweicloud.com/HuangHai/extracted/26c367a3c3cf20e62e1d6c7904f6abf6/image_1.png"
    ans = qvq_single(img_url)
    print(ans)