Bitzu

AI-901 ยท AI models and how to configure them

Choosing a model by capability

Matching model types to needs: large vs small language models, reasoning models, multimodal models, embedding models, image generation and speech models, and trade-offs in cost, latency and quality.

Key points

  • Large language models: best general quality, costlier, slower.
  • Small language models: cheaper, faster, for narrow or simple jobs.
  • Reasoning models: extra tokens on hard problems.
  • Multimodal models: accept images or audio as well as text.
  • Embedding models: text to vectors for search. Image-generation: pictures. Speech: audio to and from text.
  • Start from the task, pick the smallest model meeting your quality bar, test it.

Source: https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure

Practice questions

Single choice

Which type of model turns text into numeric vectors so you can search by meaning?

  1. Embedding model
  2. Image-generation model
  3. Text-to-speech model
  4. Chat completion model
Show answer

Answer: A. Embedding model

An embedding model outputs vectors that capture meaning. A chat model generates text, an image model draws pictures and a text-to-speech model produces audio.

Select all that apply

Which tasks require a multimodal model rather than a text-only model?

  1. Describing the contents of a photo the user uploads
  2. Answering questions about a chart in a screenshot
  3. Labeling the sentiment of a typed review
  4. Creating vectors for a library of text articles
  5. Summarizing a plain-text meeting transcript
Show answer

Answer: A. Describing the contents of a photo the user uploads, B. Answering questions about a chart in a screenshot

Multimodal models take inputs beyond text, such as images. Describing a photo and reading a chart screenshot need image input; sentiment on text, embeddings and summarizing plain text do not.

True or false

True or false: the largest model is always the right choice, because quality matters more than cost or latency.

Show answer

Answer: False

False. Larger models usually cost more and respond slower. For many tasks a smaller model meets the quality bar, so you should test and pick the smallest model that is good enough.

Practice 8 questions on this topic in Bitzu

Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.

Download on the App StoreGet it on Google Play

More AI-901 topics in this module