Bitzu

AIF-C01 ยท Designing foundation model applications

How do you choose a foundation model for an application?

Cost, modality, latency, multi-lingual support, model size and complexity, customization, input and output length and prompt caching.

Key points

  • Compare candidates against the requirements, not against a leaderboard: cost (token price), modality (text, image, audio in or out), latency, language coverage, model size and complexity, how far it can be customized, and input and output length limits.
  • Larger models are usually more capable but slower and pricier.
  • Prompt caching can cut cost and latency when a long prompt prefix repeats.

Source: https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html

Practice questions

Single choice

Which model attribute mainly determines whether a model can accept a photo together with a text question?

  1. Context window length
  2. Modality
  3. Temperature setting
  4. Price per token
Show answer

Answer: B. Modality

Modality describes which kinds of data a model accepts and produces. A multimodal model can take images and text; a text-only model cannot, regardless of size or price.

Select all that apply

Which factors directly affect the per-request cost of calling a model on demand?

  1. Number of input and output tokens
  2. Colour theme of the console
  3. Whether the response is rendered as markdown
  4. The price tier of the chosen model
  5. The length of the project name
Show answer

Answer: A. Number of input and output tokens, D. The price tier of the chosen model

On-demand pricing is token based, so the number of input tokens and output tokens drives cost, and the model's price tier sets the rate. The model's license colour or the Region name of a console tab do not change the bill.

True or false

The largest available model is always the best choice because bigger models are both more accurate and cheaper per request.

Show answer

Answer: False

False. Larger models often score higher on hard tasks but cost more per token and respond more slowly. A smaller model is frequently the better fit for simple, high-volume tasks.

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 AIF-C01 topics in this module