Meng Shao recommends Jiahui Huang's nine-part LLM training and inference video series

Meng Shao recommends a nine-episode, 230-minute YouTube series called "AI Training and Inference" by Jiahui Huang (@jbhuang0604), covering foundation model training and inference.

Shao says the series spans the full training and inference architecture, from single-GPU memory and optimizers through ZeRO, FSDP, five-dimensional parallelism, LoRA variants, speculative decoding, on-policy distillation, GRPO, AdamW and Muon.

Episode 9, LLM Serving (about 42 minutes), is member-only content. Shao says viewers can decide for themselves whether to study it after watching the first eight episodes. Huang's own post says he is still building the playlist and asks which other topics to add.