How to input multiple images using VLM-R1

[WuYeeh]

This is a very good job, but I seem to be in trouble. When I use VLM-R1 to train Qwen2.5VL for tasks which need input multiple images, the images seem cannot be input into the Trainer. LLaMA-Factory is easy to input multiple images. How should I modify the training code in VLM-R1? Looking forward to your reply, thank you.

[snakeztc]

We are working on supporting this feature. Stay tuned.

[gmyFighting]

In grpo_trainer.py, the 'image' variable is a list. You simply need to add all your images to this list.

[WuYeeh]

@gmyFighting Thanks, I get it. But I want to know if I want to place my images in the designated position of the prompt (different images in different position), can I add an identifier at the desired location like LLaMA-Factory?