Skip to main content

Exercise 2 - First API Calls

Goal: Make a real embedding call and a real generation call. Confirm both return sensible output.

Background​

Ollama exposes two endpoints you will use throughout this curriculum:

  • POST /api/embeddings - converts text to a vector
  • POST /api/generate - generates text given a prompt

Both are simple JSON over HTTP. No SDK required.

Assignment​

Open 02_first_calls.py.

Part A - Embeddings

  1. Call /api/embeddings with the model nomic-embed-text and any short sentence.
  2. Convert the returned list to a numpy float32 array.
  3. Print the shape. It should be (768,).
  4. Print the min, max, and L2 norm of the raw vector.

Part B - Generation

  1. Call /api/generate with the model llama3.2, a simple question as the prompt, and "stream": false.
  2. Print the "response" field from the JSON.
  3. Time how long the call takes using time.time(). Print the duration.

Part C - Reflect

Add a comment at the bottom of the file answering:

  • Is the embedding vector already normalised (norm ≈ 1.0)?
  • How many seconds did generation take on your machine?

Thinking questions​

  • What happens if you pass an empty string to the embedding endpoint?
  • The generation endpoint also accepts "stream": true (the default). What would you need to change in your code to handle a streamed response?

← Exercise 1 · Next: 01 Embeddings →