Support for Multimodal , multi-image summary inference

[vivekt-max]

Hi Team,

It would be great if someone has some pointers on how to achieve multimodal + mult-image inferencing using a model like QwenVL,3B model. We want to generate a summary across a set of images, using the multimodal Qwen model on-device.

Currently, it appears that the visual summary is restricted to just one image at time, using the mmproj multimodal projector, for the visual content processing.

However, --mmproj can only take one image at a time.

Can it be extended to allow multiple images ?

Kindly help!