README.md

metadata

title: RAG Khidmat (Persistent Index)
emoji: 🧠
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

RAG Khidmat — Hugging Face Space (Persistent Index)

Same as before, but now the FAISS index is saved to disk and auto-loaded on startup.

How it works

• After you click Build index (save), the app writes the index to the INDEX_DIR (default: faiss_index/).
• On app startup or when you click Load saved index, it attempts to load the saved index using the same embeddings model.
• If you upload a new PDF and rebuild, the old index folder is overwritten (by default).

Environment variables

• INDEX_DIR (default: faiss_index)
• EMBEDDINGS_MODEL (default: mixedbread-ai/mxbai-embed-large-v1)
• GENERATION_MODEL (default: google/flan-t5-base)
• HF_TOKEN (secret, only if needed)
• MAX_NEW_TOKENS (default: 256)
• TEMPERATURE (default: 0.2)