SHI-Labs/Versatile-Diffusion

Image-to-text doesn't seem to work

patrickvonplaten opened this issue · 2 comments

Hey @xingqian2018,

Great work!

I currently have problems when running the image-to-text application.

When doing the following:

and then running the following command:

python inference.py --app image-to-text --model official --fp16 --image assets/boy_and_girl.jpg --nsample 1 --seed 0

I'm getting as an output:

viewer surveys the fresh local produce

I'm running on 1.12.1+cu113 on a V100

I think the issue is on the latest dev version of the transformers library. The issue will be resolved when using transformers==4.24.0

For reference, this issue will be solved for transformers>4.24.0 as soon as this issue is resolved: huggingface/transformers#20394