Overview

LumiChats v1.1 is a 3.2 billion parameter conversational AI model developed by adityakum667388, built upon Meta's Llama 3.2 3B Instruct foundation. It has been fine-tuned using LoRA (Low-Rank Adaptation) with the Unsloth framework, enabling significant efficiency gains, including 2x faster training and inference and a reduced memory footprint through 4-bit quantization. The model was specifically trained on the mlabonne/FineTome-100k dataset with a response-only objective, focusing on generating high-quality, natural dialogue.

Key Capabilities

• Enhanced Conversational Abilities: Fine-tuned for natural and engaging dialogue, with superior instruction following.
• Efficiency: Achieves fast training and inference with Unsloth optimizations and 4-bit quantization, making it suitable for edge devices.
• Multilingual Support: Supports 8+ languages including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
• Memory Optimized: Trained with only 2.35 GB peak memory using gradient checkpointing, allowing deployment on consumer-grade hardware.
• Privacy-Focused: Capable of running entirely on-device without cloud dependencies.

Intended Use Cases

• Conversational AI: Ideal for chat applications and natural dialogue systems.
• Personal Assistants: Suitable for task management and information retrieval.
• Content Generation: Assists with creative text generation and writing.
• On-Device Applications: Optimized for mobile AI assistants and offline chatbots due to its efficiency.