DeepSeek-Harness-Vision-Tools

类型	插件	分类	视觉与多模态
星数	⭐10	状态	✅ 活跃
能力	multimodal, vision	更新时间	2026-08-13

📖 项目简介

Give your DeepSeek Harness (dsh) eyes: keep your text model as the brain, and let a local vision model do the seeing. Any text model, any vision model, on Mac or Windows/PC. The model driving dsh is usually text-only (DeepSeek, dense Qwen, Llama, Mistral). Hand it an image and dsh refuses before it sends, naming the model. Declaring input: [text, image] to force it through is worse: the endpoint then answers `400 "not a multimo

📦 安装

git clone https://github.com/tonyd2wild/DeepSeek-Harness-Vision-Tools
cd DeepSeek-Harness-Vision-Tools
cp .env.example .env      # point the endpoints at YOUR hosts
./setup.sh                # brings up a local vision server (+ the proxy with RUN_PROXY=1)

🚀 快速开始

python3 shim/vision_shim.py --port 8900 \
  --upstream http://127.0.0.1:8000 \
  --vision-url http://YOUR_FAST_VISION_HOST:8081/v1/chat/completions \
  --vision-model your-fast-vlm

🔗 链接

• GitHub 仓库
• 完整 README
• 返回DeepSeek-Harness-Vision-Tools所在分类