Large Language Model (LLM) Training

The Large Language Model (LLM) Training course gives engineers and data practitioners a thorough, end-to-end understanding of LLMs, from how they are trained to how they are served and used inside products. It sits between a conceptual introduction and a pure application course: you look under the hood far enough to make good engineering decisions, then build.

You study the transformer architecture, tokenisation, pretraining objectives, data curation, scaling behaviour, post-training with instruction tuning and preference optimisation, and the role of mixture-of-experts and long-context techniques. You then work with inference: sampling strategies, quantisation, KV caching, batching and serving open models. Application modules cover prompt and output design, retrieval-augmented generation, tool calling, evaluation, guardrails and cost control. Fine-tuning is introduced and compared with other adaptation options.

There is no external exam. You finish with a capstone and receive an Assure Success course-completion certificate. Pretraining a frontier model from scratch is not practical in a course, so labs use small models and hosted APIs.