说明

使用mindie1.0部署qwen2_vl_72b模型，可以用来分析图片了。现在想分析视频。看了下mindie文档，支持视频分析的。

错误

错误1：没安装pyav

http返回

Error code: 422 - {'error': 'Messages token length must be in (0, 1048576], but got 0', 'error_type': 'Input Validation Error'}

解决办法：安装pyav，重启mindie

错误2：使用base64

http返回：

Error code: 422 - {'error': '[InferTokenizer::DownloadUrl] Download fail: ValueError: Input type video_url does not match the mm type.', 'error_type': 'Input Validation Error'}

解决办法：不要使用base64传入，使用磁盘目录或者http下载目录

完整的例子

from openai import OpenAI
import os
# import base64
import time

#  Base64 编码格式
# def encode_video(video_path):
#     with open(video_path, "rb") as video_file:
#         return base64.b64encode(video_file.read()).decode("utf-8")

# 将xxxx/test.mp4替换为你本地视频的绝对路径
# base64_video = encode_video("./数据集上传.mp4")
client = OpenAI(
    api_key="sk-suibianxie",
    base_url="http://10.20.xx.xx:port/v1",
)
start_time = time.time()  # 记录开始时间
completion = client.chat.completions.create(
    model="qwen_1_0c081946",
    messages=[
        {
            "role": "user",
            "content": [{"type":"text","text": "You are a helpful assistant."}]
        },
        {
            "role": "user",
            "content": [
                {
                    # 直接传入视频文件时，请将type的值设置为video_url
                    "type": "video_url",
                    #"video_url": {"url": f"/in/update.mp4"}, 
                    "video_url": {"url": f"http://10.20.xx.xx:9010/%E6%95%B0%E6%8D%AE%E9%9B%86%E4%B8%8A%E4%BC%A0.mp4"},
                    # "fps": 1.0,
                    # "max_pixels": 360 * 420,
                },
                {"type": "text", "text": "这段视频描绘的是什么景象?"},
            ],
        }
    ],
)
end_time = time.time()  # 记录结束时间
print(f"请求耗时: {end_time - start_time:.2f}秒")  # 打印耗时
print(completion.choices[0].message.content)

参考资料

使用python访问mindie部署的vl多模态模型