Multi-Image Input Problem

[hexmSeeU]

I tried to input a sequence of images to the Owl3 7B model in the following format:

I input 7 images into the model and asked how many images it could see. It answers "1". How can I solve that?

[Michael4933]

Well I encountered the same issue, and I highly doubted the model's original capacity

[Fu-Fu-Fu-Fu]

I also found that when I input multiple photos and asked them to describe the content of the photos separately, the number of photos output by the model did not match and sometimes said nonsense. Have you solved this problem?

[hexmSeeU]

I also found that when I input multiple photos and asked them to describe the content of the photos separately, the number of photos output by the model did not match and sometimes said nonsense. Have you solved this problem?

Not yet. However, I have tried to use another framework [ms-swift] to run the inference/training code of MLLM, which also incorporates mPLUG-owl series.