Multi-GPU AI Train the Trainer

The Multi‑GPU AI Train‑the‑Trainer Workshop is a fully online, five‑day programme designed to equip instructors with both the technical expertise and pedagogical strategies needed to teach modern multi‑GPU and distributed AI workflows. Developed collaboratively within CASTIEL2 WP3 together with NCCs from Finland, Hungary, the Netherlands, Belgium, Romania, Italy, and Sweden, the workshop builds on the CASTIEL2 Multi‑GPU AI curriculum and provides hands‑on training using CINECA’s Leonardo supercomputer. Participants explore a broad range of topics—including GPU architectures, deep learning fundamentals, distributed training with PyTorch DDP, model and pipeline parallelism, PyTorch Lightning, LLM fine‑tuning with HuggingFace Accelerate and DeepSpeed, computer vision, MLOps on HPC, Ray for distributed AI, RAG workflows, and hyperparameter optimisation—while also gaining practical guidance and reusable materials to confidently deliver these trainings within their own national programmes.