Beam Search & Constrained Decoding
Theory, math, and algorithms of beam search and constrained decoding for LLM generation.
Theory, math, and algorithms of beam search and constrained decoding for LLM generation.
How graph algorithms power knowledge graphs and GraphRAG retrieval in production LLM systems.
BPE, WordPiece, and Unigram LM — the subword tokenizers that power virtually every modern LLM.