environment.yml

# Conda environment for SLAE (autoencoder loading + inference).
#
# After creating the environment, install the PyTorch Geometric companion
# wheels that match your torch + CUDA build, then install SLAE itself:
#
#   conda env create -f environment.yml
#   conda activate slae
#   pip install torch-scatter torch-cluster torch-sparse \
#       -f https://data.pyg.org/whl/torch-2.6.0+cu124.html
#   pip install -e .
#
name: slae
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.12
  - pip
  # structural-biology stack that installs more cleanly from conda-forge
  - conda-forge::biotite
  - conda-forge::rdkit
  - conda-forge::pandas
  - conda-forge::pyarrow
  - pip:
      - torch>=2.2
      - torch-geometric==2.7.0   # 2.8 needs pyg-lib for radius_graph; 2.7 uses torch_cluster
      - torch-runstats
      - e3nn>=0.5
      - einops
      - einx
      - opt-einsum-fx
      - esm>=3.0
      - hydra-core>=1.3
      - omegaconf>=2.3
      - lightning>=2.2
      - "numpy<2"
      - biopandas
      - atomworks
      - loguru
      - tqdm
      - pyyaml
      - packaging
      # NOTE: install torch-scatter / torch-cluster / torch-sparse from the
      # matching PyG wheel index (see comment above) -- they are not pulled
      # here because the correct wheel depends on your CUDA version.