Requirements.txt - wrong versions

[google-deepmind]

Requirements.txt - wrong versions

[Chirann]

I've got the same error when I'm installing requirements under python 3.10.13.

[p-i-]

(Ubuntu 20.04) Using conda create -n geom python=3.10 I get python 3.10.13 and pip install -r requirements.txt works.

[Westlifers]

I think this may be caused by the fact that jaxlib doesn't support Windows until version 0.4.13?(check here) I'm on Windows and tried several versions of Python, all failed at this step. Manually install higher version and delete jaxlib from requirements.txt may help you. (However there are more version problems than I expected and I haven't fixed them all, maybe just switch to linux is the best way)

[Landau1994]

This is my installation step in wsl2. You can trys this. I already create conda envrionment python==3.10.9

install required package

This is following the guidence of the auothor,but you need install the following packages first

# download from pypi

pip install /mnt/d/tmp/tfds_nightly-4.9.2.dev202308090034-py3-none-any.whl

pip install /mnt/d/tmp/seqio_nightly-0.0.17.dev20231013-py3-none-any.whl

# install from git directory

pip install --require-hashes -r requirements.txt

enalbe gpu

install cuda in conda environment

This is required for tensorflow and jax

conda install cuda -c nvidia/label/cuda-11.8.0

conda install -c conda-forge cudnn=8.4.1

pip install nvidia-cudnn-cu11==8.6.0.163

export LD_LIBRARY_PATH

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/wlt2023/anaconda3/envs/alphageometry/lib:/home/wlt2023/anaconda3/envs/alphageometry/lib/python3.10/site-packages/nvidia/cudnn/lib

Then install

install jaxlib

First

pip install jax==0.4.6 jaxlib==0.4.6 orbax-checkpoint

Then choose the correct jaxlib from https://storage.googleapis.com/jax-releases/jax_cuda_releases.html, and donwload the correct version of jaxlib

pip install /mnt/d/tmp/jaxlib-0.4.6+cuda11.cudnn82-cp310-cp310-manylinux2014_x86_64.whl

run

you should add this to avoid matplotlib error

export DISPLAY=:0

[tpgh24]

I had the same issue and found a workaround by using --no-deps option. I made some improvements in a fork repository and have some ideas to further improve the performance of AlphaGeometry, check out AG4Masses and issue 110. It has detailed instruction on set up.