Training systems

Data, tensor, pipeline, and expert parallelism. Checkpointing, elasticity, and the difference between a training recipe and a training system.

Notes coming.