An animated, self-contained guide to LLM inference parallelism: DP, TP, PP, EP, their combinations, and the attention designs (MHA, GQA, MQA, MLA) that make them work.
22 October 2026