Week 5 ML Series: How Tensors Power Every AI Model, From CPU to GPU

A developer progressing through a 32-week machine learning curriculum reached Week 5, focusing on tensors — the core data structure underlying all AI models. Tensors are typed, multi-dimensional arrays that can reside on either a CPU or GPU, and are described by their rank, shape, dtype, and device. The session introduced scalars, vectors, matrices, and higher-dimensional tensors using PyTorch, with hands-on scripts demonstrating element-wise operations, broadcasting, and CPU-to-GPU memory transfers. All experiments were run remotely via SSH on an NVIDIA DGX Spark machine using a pre-configured PyTorch environment with CUDA support. This week also marked the start of a parallel CUDA track within the broader goal of understanding how machine learning inference works at a low level.