duhai-vision0 ★@hamliy-fengVisual model adapter for Codex and DeepSeek Harness, powered by PaddleOCR-VL and Qwen.NewVision