Hi, I tried to run the attention map diffusers for StableDiffusion3 pipeline, basically everything when fine when I simply inference with this SD3 model. However, when I call the init_pipeline() from your utils, the inference speed slows down and when it reached the 18th(or above) of 28, the inference is killed unexpectedly without any warning or error. Resulting neither denoised image or attention map is outputed.
Looking forward to your reply, thanks!