Engrain by ByteRush

AGT.02 7 concepts

LLM fundamentals

Explain how LLMs work well enough to make engineering decisions.

  1. AGT.02.01Tokens, context windows and next-token predictionCoreJuniorBeing written
  2. AGT.02.02Transformers and attention (intuition)Mid-levelBeing written
  3. AGT.02.03Training: pre-training, post-training and reasoning modelsCoreMid-levelBeing written
  4. AGT.02.04Generation controls, model choice and costCoreJuniorBeing written
  5. AGT.02.05Embeddings and similarityCoreJuniorBeing written
  6. AGT.02.06How inference serving worksMid-levelBeing written
  7. AGT.02.07ML basics for AI engineersJuniorBeing written

Interview questions in this module

  1. What is a token? Why does token count matter?
  2. Why do LLMs hallucinate? How do you reduce it in a product?
  3. Explain attention in simple terms.
  4. What is the difference between pre-training and fine-tuning?
  5. How do you choose between a small and a large model for a feature?
  6. What is an embedding and how is it used in search?
  7. What are reasoning models and when are they worth the extra cost?
  8. What is the KV cache and why does long context cost more?
  9. Explain precision and recall with an example.