LLaMA-3 GenZ Chatbot with Google Search Integration

This project features a fine-tuned LLaMA-3-8B model that mimics Gen-Z language and rephrases real-time Google search results into casual, humorous, and stylistic responses.

The chatbot uses live search via the Serper API and responds in a Gen-Z tone. It demonstrates real-world application of retrieval-augmented generation (RAG), language style transfer, and optimized inference on large language models.

Features

• Fine-tuned LLaMA-3-8B model using LoRA for Gen-Z conversational tone
• Integrated with Serper API to fetch live Google search results
• FastAPI-based backend for low-latency REST API inference
• 4-bit quantized model using bitsandbytes for reduced memory usage and faster response
• Contextual prompt chaining and dynamic response generation

Tech Stack

• LLaMA-3-8B (fine-tuned with LoRA adapters)
• Quantization via bitsandbytes (4-bit inference)
• FastAPI backend
• Serper API (Google Search integration)
• Hugging Face Transformers
• Prompt engineering for language style transformation

Example

Input Query:
"Why is the sky blue?"

Fetched Result (via Serper):
"Blue light from the sun scatters in all directions by the gases and particles in the Earth's atmosphere."

Model Output:
"So basically, the sky filters vibes and blue just wins. Physics but make it aesthetic."

Model & Weights

This chatbot uses a LoRA adapter fine-tuned on top of the base meta-llama/Meta-Llama-3-8B model.

The adapter is available on Hugging Face:

Adapter Model: notninja/chad-gpt

Credits

Developed by:

• Ved Thorat
  GitHub: i3hz

• Pushkar Sharma GitHub: pushkar-hue