Fine-tuning is a specialization process where a pre-
trained AI model is refined with
training using specific data for a concrete task. It's like giving specialized
training to a model that already has general knowledge.
During fine-tuning, a model that has already been
trained with large amounts of general data is adapted for a specific use using a smaller, specialized
dataset. For example, you can take a
large language model trained with general texts and refine it with medical texts to specialize in medical terminology and concepts.
This process is more efficient than
training a model from scratch, as it leverages
knowledge transfer from the model's previous general learning and adds specific knowledge. Imagine a model
trained to recognize objects in images that can be refined with photos of birds to become an expert in identifying bird species, without having to learn basic concepts like shapes, colors, or textures from scratch.
Unlike other
AI model specialization techniques such as
LoRA, which adds adaptive layers, fine-tuning modifies the complete model to incorporate new specific knowledge. This requires more computational capacity and
training data.