GPU Training Lab

Overview

The GPU Training Lab is where you move from CPU-bound ML experiments to real deep learning. Each student gets a dedicated NVIDIA T4 GPU with 16 GB of VRAM — enough to train ResNets, fine-tune BERT-sized models, run object detection pipelines, and experiment with generative models. The T4 is the workhorse GPU of cloud AI: it supports FP16 mixed-precision training, INT8 inference, and is the same hardware used in production at many companies. You'll use this lab for the majority of exercises in intermediate and advanced courses where training speed and GPU memory matter.

What You'll Do in This Lab

• Train convolutional neural networks (ResNet, EfficientNet, YOLOv8) from scratch
• Fine-tune pretrained models from Hugging Face for custom tasks
• Run mixed-precision training with PyTorch AMP for faster iteration
• Track experiments with Weights & Biases — log metrics, compare runs, visualize training curves
• Profile GPU utilization with nvidia-smi and PyTorch Profiler
• Export trained models to ONNX format for deployment

Lab Workflow