Is \n behind the <image> or <video> token needed?

[pmtheman]

Thanks for your work!
In the code from the official repository, there is no \n behind the token.
I don't know whether it is a requirement?

[2U1]

Yes for the consistency with my other repository, it is needed after the <image> token or <video> token.
Or you could just modify the lines here.

Qwen2-VL-Finetune/src/training/data.py

Lines 325 to 329 in 98407a7

	if is_video:
	input_string = input_string.replace(LLAVA_VIDEO_TOKEN+'\n', VISION_START_TOKEN+DEFAULT_VIDEO_TOKEN+VISION_END_TOKEN)
	else:
	input_string = input_string.replace(LLAVA_IMAGE_TOKEN+'\n', VISION_START_TOKEN+DEFAULT_IMAGE_TOKEN+VISION_END_TOKEN)

[2U1]

I've updated the code to handle if there are no "\n" behind the <image> or <video> token.