A newer version of the Streamlit SDK is available:
1.54.0
title: DocTalk - Chat With PDF
emoji: 📗💬
colorFrom: indigo
colorTo: pink
sdk: streamlit
sdk_version: 1.35.0
app_file: app.py
pinned: false
📗💬 DocTalk - Chat With PDF
An intelligent, completely free-to-run PDF chat application powered by Google's Gemma-2-2b-it model. Optimized for CPU usage on Hugging Face Spaces.
✨ Features
🤖 Core Engine
- Model: Google Gemma-2-2B-IT (Instruction Tuned)
- Architecture: Runs entirely locally on CPU (no GPU required)
- Performance: Optimized with FAISS for instant vector retrieval
🎯 Key Capabilities
- ⚡ CPU Optimized - Runs smoothly on Hugging Face Free Tier
- 📤 Easy Upload - Simple sidebar PDF upload
- 🧠 Smart Context - Uses
all-MiniLM-L6-v2for precise semantic search - 💬 Memory - Maintains chat history within the session
- 🔒 Secure - Handles Hugging Face tokens via environment secrets
🚀 How to Use
1. Set Up Authentication
- This app requires a Hugging Face Access Token (Read permissions) to download the Gemma model.
- For Users: Enter your token in the app sidebar if prompted (or set it in Space secrets).
2. Upload Your PDF
- Navigate to the sidebar
- Click "Browse files" to upload your PDF document
- Click "🚀 Process Document"
3. Start Chatting!
- Wait for the "✅ Ready to chat!" notification
- Type your question in the chat input at the bottom
- Receive concise, context-aware answers from Gemma-2
🛠️ Technical Stack
- Frontend: Streamlit
- LLM: google/gemma-2-2b-it
- Embeddings: sentence-transformers/all-MiniLM-L6-v2
- Vector Store: FAISS (Facebook AI Similarity Search)
- PDF Processing: PyPDFLoader
- Orchestration: LangChain
📦 Installation (Local)
To run this app on your own machine:
https://huggingface.co/spaces/ChiragKaushikCK/Chat_with_PDF
🌟 Features Breakdown FAISS Vector Search Replaces heavy database lookups with lightweight, in-memory similarity search.
Ensures responses are strictly grounded in your uploaded document.
Pre-loaded Models The embedding models are cached (@st.cache_resource) to ensure the app feels snappy after the initial cold start.
Gemma-2-2B-IT Google's latest lightweight open model.
Instruction-tuned for better Q&A performance compared to base models.
Small enough (~2.6B params) to fit in standard RAM.
⚠️ Limitations Speed: Since this runs on CPU, generating long answers may take a few seconds.
Memory: Designed for standard PDFs. Extremely large files (500+ pages) might hit RAM limits on free tiers.
Session: Chat history is cleared if the page is refreshed.
🤝 Contributing Contributions are welcome! Please feel free to submit issues or pull requests to improve the UI or add new features.
📄 License MIT License
🔗 Links Google Gemma Models
LangChain Documentation
Streamlit