environment-pipeline.yml

# Conda environment for reproducing PluRule end-to-end (pipeline/).
#
# Setup:
#   conda env create -f environment-pipeline.yml
#   conda activate plurule-pipeline
#
# Requires:
#   - Multiple CUDA GPUs (8B embedding model, 30B LLM judge)
#   - Reddit API keys in credentials/reddit_api_keys.json  (Stage 2)

name: plurule-pipeline
channels:
  - conda-forge
dependencies:
  - python>=3.10
  - aria2
  - pip
  - pip:
      # I/O (shared with hydrate)
      - zstandard>=0.21.0
      - orjson>=3.9.0
      - tqdm>=4.65.0
      - requests>=2.31.0
      - torf>=4.2.0

      # Numeric / plotting
      - numpy>=1.24.0
      - pandas>=2.0.0
      - scipy>=1.10.0
      - scikit-learn>=1.3.0
      - matplotlib>=3.7.0

      # Embedding matcher (Stage 3, 9a) + LLM judge (Stage 8) + cluster labeler (Stage 9c)
      - torch>=2.0.0
      - transformers>=4.40.0
      - vllm>=0.6.0

      # Clustering (Stage 9b)
      - umap-learn>=0.5.5
      - hdbscan>=0.8.33

      # Reddit API (Stage 2)
      - praw>=7.7.0