Glossary

• LLM: Large Language Model.
• RAG: Retrieval-Augmented Generation; combines retrieval with generation.
• Token: Unit of text used by tokenizers; affects cost and limits.
• Quantization: Reducing numerical precision (e.g., FP16→INT4) to speed/fit.
• Beam Search: Decoding strategy exploring multiple candidates.
• KV Cache: Cached key/value tensors to speed autoregressive decoding.
• Guardrails: Controls to steer or constrain model outputs.