Exploring LLMs as a Powerful AI Engine
In the previous chapter, we saw the structure of a transformer, how it is trained, and what makes it so powerful. The transformer is the seed of this revolution in natural language processing (NLP), and today’s large language models (LLMs) are all based on transformers trained at scale. In this chapter, we will see what happens when we train huge transformers (more than 100 billion parameters) with giant datasets. We will focus on how to enable this training at scale, how to fine-tune similar modern ones, how to get more manageable models, and how to extend them to multimodal data. At the same time, we will also see what the limitations of these models are and what techniques are used to try to overcome these limitations.
In this chapter, we'll be covering the following topics:
- Discovering the evolution of LLMs
- Instruction tuning, fine-tuning, and alignment
- Exploring smaller and more efficient LLMs
- Exploring...
