Model

The model combines a supervised learning component with an unsupervised learning component. The supervised learning component processes the dataset and feeds this to a base model, while the unsupervised component leverages diverse data sources to further enhance the model’s capabilities. These learned weights are then combined with the pre-trained base model.

For the fine-tuning process, the base model is optimized using the Low-Rank Adaptation (LoRA) technique. By applying this method, the model’s performance is refined, enabling it to better adapt to specific tasks. In this experiment, the Llama 3.2 3B model serves as the foundational base for the supervised and unsupervised learning.