Feature/faiss index caching

[ahuou]

🚀 Add Optional FAISS Index Caching to RAG Evaluations

📌 Summary

This PR adds the ability to reuse or save FAISS indexes in RAG mode, reducing repeated embedding and chunking overhead when running multiple evaluations on the same dataset.

✅ Key Features

• New config option: faiss_index_path under rag_params
• Automatically:
  • Loads existing index if available
  • Saves new index to disk for reuse
• Fully backward compatible — if no path is provided, behavior remains unchanged

🔧 Modified Files

• main_accelerate.py: Passes faiss_index_path from YAML
• embed_model.py: Builds or loads FAISS index based on the given path

🧪 Tested

• Run once: builds & saves FAISS index
• Run again: loads index successfully and skips rebuild

📁 Example rag_model.yaml

rag_params:
  embedding_model: "BAAI/bge-base-en"
  docs_name_or_path: "your_dataset"
  top_k: 5
  num_chunks: 100000
  similarity_fn: "cosine"
  faiss_index_path: "indexes/my_index"