# EmpathRAG — Full development / training dependencies # # Install order matters — read INSTALL NOTES below before running pip install. # The slim runtime requirements live in requirements.txt and are what the # Hugging Face Space and Gradio demo actually import. This file adds the # baseline-iteration training stack (RoBERTa, DeBERTa, Mistral, FAISS) used # by the notebooks and the V1 pipeline. # # ── INSTALL NOTES ────────────────────────────────────────────────────────────── # 1. PyTorch (CUDA) - install FIRST before anything else: # pip install torch==2.5.1+cu121 --index-url https://download.pytorch.org/whl/cu121 # # 2. llama-cpp-python (CUDA wheel) - install SECOND: # pip install llama_cpp_python==0.3.4 --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121 # # 3. Everything else: # pip install -r requirements-dev.txt # # 4. Force correct numpy (faiss requires <2.0): # pip install "numpy==1.26.4" --force-reinstall # ────────────────────────────────────────────────────────────────────────────── # Slim runtime (also pulled by requirements.txt) -r requirements.txt # HuggingFace stack — baseline iteration only transformers==4.57.6 peft==0.18.1 datasets==2.18.0 accelerate==1.13.0 evaluate==0.4.1 # Retrieval — baseline iteration only faiss-cpu==1.9.0.post1 sentence-transformers==3.4.1 # Explainability — baseline iteration only captum==0.7.0 # Evaluation ragas==0.1.21 bert-score==0.3.13 rank-bm25==0.2.2 # Data utilities only used by training notebooks kaggle==1.6.3