Introduction to Low Rank Adaptation and Quantization
To begin, let's understand the concept of LORA in AI models. LORA refers to a technique known as low rank adaptation. Imagine you have a giant box of Legos with which you can build various things like cars and spaceships. However, this giant box is heavy and not very portable. Similarly, a large language model, such as GPT-4, is powerful but computationally demanding.
To address this, low rank adaptation comes into play. It involves creating a smaller and lighter version of the large language model that is specifically adapted for a particular task.
