Skip to content

Foundation Model

  • Foundation Models are trained on a wide variety of input data (unlabeled data)
  • Foundation models may cost tens of millions of dollars to train
  • FMs can be fine-tuned if necessary to better fit the use-case

Generative AI

Generative modality

  • Modalities
  • Text
  • Image
  • Vision

  • Embedding

  • Some models are multimodal

Large Language Models (LLM)

  • Text To Text
  • Relies on a foundation model
  • Designed to generate coherent human-like text
  • We usually interact with LLM via a prompt

  • The generated text is non-deterministic

  • Context Window

  • The number of tokens an LLM can consider when generating text (the "size" of your prompt)
  • Large context windows require more memory and processing power Context Window

Diffusion Models

  • Text To Image
  • Trains using a forward diffusion process
  • Adds noise to the image
  • Generates image from noise
  • Reverse the diffusion process

Diffusion