Fix timm version

For me, fixing the timm version seems to suffice to fix #242 locally. Other than that I kept the packages that are installed by `pip install`; for the package versions mentioned in https://github.com/clovaai/donut/#software-installation this means ``` torch 2.0.1 torchvision 0.15.2 pytorch-lightning 2.0.9 transformers 4.33.2 ``` Notice that I only tested the cpu version and the demo in `app.py`, but did not attempt any training.