What You Will Learn

• Build and train deep learning models using PyTorch 2.x across NVIDIA GPUs (CUDA), Apple Silicon (MPS), and CPUs.
• Master tensor calculus, memory strides, contiguous layouts, and Autograd automatic differentiation.
• Implement modular neural networks with custom layers, activations (ReLU, GELU), and weight initializations.
• Construct production training loops with AdamW, learning rate scheduling (OneCycleLR, Cosine), and gradient clipping.
• Build resilient data pipelines with custom Datasets, multi-worker DataLoaders, and torchvision transforms v2.
• Architect Convolutional Neural Networks (CNNs) and fine-tune modern Vision Transformers (ViT, ConvNeXt).
• Deconstruct Transformer internals: Multi-Head Self-Attention, RoPE position embeddings, GQA, and FlashAttention.
• Synthesize novel data using Generative AI: Variational Autoencoders (VAEs), GANs, and Latent Diffusion models.
• Fine-tune Large Language Models (LLMs) with Parameter-Efficient Fine-Tuning (PEFT, LoRA & 4-bit QLoRA).
• Scale training across multi-GPU clusters using DistributedDataParallel (DDP) and Fully Sharded Data Parallel (FSDP2).
• Export models for production serving using modern torch.export, ONNX Runtime (dynamo=True), ExecuTorch, and FastAPI.

Part 1: Foundations & Core Mechanics — Chapters 1 to 4

Master the fundamental mathematical abstractions, tensor geometry, and automatic differentiation mechanics powering PyTorch.

Part 2: Training, Data Pipelines & Computer Vision — Chapters 5 to 8

Learn how neural networks optimize, ingest massive datasets, evaluate performance, and classify images.

Part 3: Transformers, Generative AI & LLM Engineering — Chapters 9 to 12

Deep-dive into modern AI architectures: fine-tune Vision Transformers, construct attention from scratch, generate images with diffusion, and fine-tune LLMs with LoRA.

Part 4: Scaling, Export & Production Deployment — Chapters 13 to 16

Benchmark and profile bottlenecks, scale training across multi-GPU clusters, export models with modern torch.export, and deploy production APIs.

Course Details

• Target Technologies: PyTorch 2.x, TorchVision, Hugging Face PEFT, ONNX Runtime, ExecuTorch, and FastAPI.
• Hardware Support: NVIDIA GPU (CUDA), Apple Silicon (MPS Metal), and Standard CPU.
• Format: 16 self-contained modular chapters with practical implementation code, architecture diagrams, and exercises.
• Prerequisites: Python programming knowledge and basic linear algebra concepts.

Click on Chapter 1: Getting Started with PyTorch above to begin.