Batch Inference Produces Incorrect Results

[jiangzeyinzi]

Thanks to the authors for the excellent work.

During the execution of the code, when I use a prompt list as input, i.e., batch inference, it results in incorrect outcomes. I suspect this is likely related to position encoding, but I have not pinpointed the exact cause. Could you please help verify the batch inference? Additionally, when setting --num_images_per_prompt > 1, there is also an STG error occurring. Thank you.