Search for a command to run...
Wo Vision zu Text wird: Lokalisierung des OCR-Routing-Bottlenecks in Vision-Language Models