Does it support multiple images per prompt?

In your sample code, it only takes a single image path as input. I tried to modify the model code to support multiple images, but failed. Even though the code seems to support multi-image processing. Therefore, I’d like to ask whether DeepSeek-OCR works like Qwen-VL 2, which supports multiple images as input for cross-image reasoning. Or I have to concat multiple images into one as the input?