search

TencentARC / SEED-Story

SEED-Story: Multimodal Long Story Generation with Large Language Model
https://arxiv.org/abs/2407.08683
Other
734 stars 56 forks source link

Segmentation Fault. #11

Closed floppycracken closed 3 months ago

floppycracken commented 3 months ago

When I run the inference file: gen_george.py I get a segmentation fault; the fault occurs when executing the code:images_gen = adapter.generate(image_embeds=output['img_gen_feat'], num_inference_steps=50)

I think some issue with the diffusion detokenizer, do you have any what might be going wrong?

floppycracken commented 3 months ago

I looked at where the issue originated; it seems it is from the adapter_modules file when the function self.sdxl_pipe() is called. It could be due to the different diffuser versions. Can you provide the version you used?

AndysonYs commented 3 months ago

Hi, I use sdxl 1.0(https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0).

floppycracken commented 3 months ago

Thank you for the reply. I was able to sort out the issue. For anyone else with a similar problem, diffusers version 0.25.1 with Cuda version 12.2 worked for me.