Inference Problem

[URRealHero]

As I mentioned in issue #18 The anole often generate nothing (Only an token). I'm not sure if there is any special token that conflict with the input? or there is more problem in multimodal-in multimodal-out

[URRealHero]

Still really sorry to bother you, It seems that '<' and '>' will lead to no generation.
However, after removing them, the model still refuse to generate images, most of the time, it only generate two images
image1.png and image3.png
I have no idea why it is
Moreover, the inference time is really long when generating image, Any ideas?