cotignacMafia

A Python test suite for evaluating video description capabilities using the Qwen3-VL-32B-Thinking model.

Description

This project tests the Qwen VL (Vision-Language) model's ability to precisely describe and analyze video content. The test script encodes local video files and sends them to the Qwen API for detailed analysis.

Features

• Video encoding to base64 for API compatibility
• Streaming response output
• Detailed video content analysis
• Product demonstration detection and timestamping
• Visual quality and content potential scoring

Prerequisites

• Python 3.8+
• DashScope API Key (from Alibaba Cloud Model Studio)

Installation

1. Clone the repository:

git clone https://github.com/ZeiProX76/cotignacMafia.git
cd cotignacMafia

2. Create and activate a virtual environment:

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Set up your environment variables:

cp .env.example .env
# Edit .env and add your DASHSCOPE_API_KEY

Usage

Run the video description test:

python test_video_description.py

The script will:

1. Load your video file
2. Encode it to base64
3. Send it to the Qwen3-VL model
4. Stream and display the detailed analysis

Configuration

Edit the video_path variable in test_video_description.py to point to your video file:

video_path = "/path/to/your/video.mp4"

API Documentation

This project uses the Qwen API in OpenAI-compatible mode:

• Base URL: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
• Model: qwen3-vl-32b-thinking

License

MIT

Author

ZeiProX76