CRIA v1.3: A Fine-Tuned Llama 2 Model for Chatbot Development

CRIA v1.3 is a 7 billion parameter language model developed by davzoku, built upon the llama-2-7b-chat-hf architecture. The project's core objective is to provide a comprehensive, end-to-end chatbot system, from model instruction-tuning to web deployment.

Key Capabilities & Training Details

• Instruction-Tuned: Fine-tuned using QLoRA (4-bit precision) to enhance conversational abilities.
• Dataset: Training was conducted on the mlabonne/CodeLlama-2-20k dataset, suggesting a potential emphasis on code-related or technical dialogue.
• Quantization: Utilizes bitsandbytes for 4-bit quantization (nf4 type) during training, enabling efficient resource usage.
• Deployment Focus: Designed to serve as the foundational model for the CRIA chat platform and integrate with web frameworks like Next.js.

Good For

• Developers building custom chatbot applications requiring a fine-tuned Llama 2 base.
• Projects focused on instruction-following and conversational AI, particularly those benefiting from a code-centric training dataset.
• Experimentation with QLoRA fine-tuning and efficient model deployment.