Efficient Data Loading Best Practice 1. Use num workers = min(8, os.cpu count()) for DataLoader 2. Enable pin memory=True when using GPU 3. Use persistent workers=True to avoid re-spawning 4. Pre-compute and cache transformations when possible 5. For image data: use torchvision.transforms.v2 (faster) 6. For large datasets: consider memory-mapped files or WebDataset 7. Profile with torch.utils.bottleneck to find I/O bottlenecks ---