Reviewing generative AI system design

The case studies connect training choices with the services that deliver model outputs. Use them to explain how input and output shape change the data pipeline, capacity estimate, serving path, and evaluation plan. Their numerical workloads are planning examples; a model name or peak hardware rate does not turn them into benchmarks.

Compare the case studies

The text case uses Llama 3.2 3B as a reference for data preparation, conversational context, model serving, and response delivery. Trace how authorized retrieved text becomes input tokens, and keep that request path separate from offline training and model release.