🤖 AnswerGPT

AnswerGPT is a real-time, context-aware, hybrid chatbot powered by:

• 🧠 Phi-2 — A powerful, lightweight language model by Microsoft
• 📦 QLoRA Fine-Tuning — Trained on your own interactions
• 💾 Qdrant Vector DB — For memory storage and semantic search
• 🌐 DuckDuckGo Web Scraping — To pull live, real-world info
• 🎛️ Gradio Web UI — For a slick ChatGPT-style interface

---

🚀 Features

✅ Fine-tuned local model (Phi-2 + QLoRA) ✅ Real-time web + memory hybrid answering ✅ Custom prompt formatting ✅ Add long-term memory via remember command ✅ Local Gradio UI ✅ Modular and hackable Python code

---

🧱 Project Structure

answergpt/
├── src/
│   ├── main.py                  # CLI chatbot
│   ├── model_loader.py          # Load base + LoRA model
│   ├── prompts.py               # Prompt formatting
│   ├── modules/
│   │   ├── vector_store.py      # Qdrant memory interface
│   │   ├── web_search.py        # DuckDuckGo scraping
│   │   └── answergpt-qlora/     # Fine-tuned LoRA model folder
├── gradio_ui.py                # Web UI with Gradio
├── data/
│   └── log.jsonl                # Chat logs for future training
├── models/                     # [Optional] model artifacts
├── .gitignore
└── README.md

---

🛠️ Setup (Local)

1. Clone the Repo

git clone https://github.com/AshrithChandan/Answer_GPT.git
cd answergpt

2. (Optional) Create Virtual Environment

python -m venv venv
venv\Scripts\activate  # On Windows

3. Install Dependencies

pip install -r requirements.txt

✅ Or install manually:

pip install transformers peft qdrant-client sentence-transformers colorama gradio duckduckgo-search

4. Put Your Fine-Tuned Model in:

models/answergpt-qlora/

This should contain:

• adapter_model.safetensors
• adapter_config.json
• tokenizer_config.json, etc.

5. Run CLI Chatbot

python -m src.main

6. Run Gradio UI

python gradio_ui.py

It will launch in your browser at http://127.0.0.1:7860

---

✨ Commands

• remember ... — Adds a fact to long-term vector memory (Qdrant)

You: remember Pluto is a dwarf planet

• ask anything — Hybrid context-aware answer

You: What is Pluto?

---

🧠 Training Your Own Model

You can fine-tune Phi-2 with QLoRA using the built-in training pipeline on Colab. Just run:

src/modules/finetune_qlora.py

Make sure data/log.jsonl has high-quality Q&A logs.

---

🧪 Try It in Google Colab

You can fine-tune or run AnswerGPT directly in Colab:

👉 Open in Colab

---

🌍 Credits

• Phi-2 by Microsoft
• QLoRA
• Gradio
• Qdrant

---

📄 License

This project is for educational & research purposes. You are responsible for compliance with licenses of pretrained models.

---

💬 Contact

Feel free to reach out via GitHub Issues or contribute with a PR! Let's build the future of local AI together 🚀