If I understand the image processing part [1] of the code correctly, each image is scaled down (max 448x224) and only uses <= 512 tokens. 14x14 pixels is one in

If I understand the image processing part [1] of the code correctly, each image is scaled down (max 448x224) and only uses <= 512 tokens. 14x14 pixels is one input token. [1] https://github.com/haotian-liu/LLaVA/blob/ee175feb646f4673fe1d6410fe556695ff26af35/llava/serve/model_worker.py#L194-L205 https://github.com/haotian-liu/LLaVA/blob/ee175feb646f4673fe...