Hands-on notes from building an ML training and serving platform: distributed training (PyTorch DDP) and Kubernetes. Each lab starts from a question, runs on real code, and explains what the result means.

Interactive labs

ML platform interactive labs: step through the PyTorch DataLoader loop, compare training with and without DistributedSampler, see where time goes with persistent_workers, and delete or crash Pods behind a Kubernetes Service.

Flashcards

ML platform flashcards: tap-to-reveal cards on DDP, data loading, Kubernetes, and system design answers, written as what / how / so what.

Code

cpp_refresh on GitHub: the experiments behind these pages (DDP benchmarks, Kubernetes labs, C++).