View on GitHub

AI in HPE

An open-source, open-access resource on generative AI in health professions education.

Large language models

A large language model is a language model with many parameters (typically billions of weights), trained on large quantities of unlabeled text, using self-supervised machine learning (or semi-supervised learning).

https://youtu.be/ZRf2BfDLlIA?si=RGv3CPKAGOcbGpkO

https://youtu.be/wbGKfAPlZVA?si=4vvKxNkEr7vHUXTY

Types of language models

Language models can be categorised in several different ways;

  1. License
  2. Size

License

Open-source language models e.g. Llama (Meta), and Mixtral (Mistral).

Commercial language models e.g. GPT (OpenAI), and Gemini (Google).

Size

Language models come in a wide range of sizes. For example, in 2024 LLaMA 2 was available in 7 billion, 13 billion, and 70 billion parameter sizes. Smaller models are cheaper to deploy and run; larger models are more capable (Facebook research, 2024).

Deployment

Large language models are deployed and accessed in a variety of ways, including:

  1. Self-hosted: Using local hardware to run inference. For example, running Llama 2 on a powerful desktop or laptop using an open-source LLM. This is the best option for privacy/security or if you already have a GPU.
  2. Cloud hosted: Using a cloud service provider to deploy an instance that hosts a specific model. For example, running an open-source model on cloud providers like AWS, or Azure. This is the best approach for customising models and their runtime (e.g. fine-tuning a model for your use case).
  3. Hosted API: Call LLMs directly via an API. There are many companies that provide language model inference APIs including AWS Bedrock, Replicate, Anyscale, Together and others. This is the easiest option overall.

Additional reading