Dataset Pipeline Builder AI Prompt

Build a production-quality data loading pipeline for training a {{model_type}} model on {{data_format}} data.

Requirements:
1. PyTorch Dataset class:
   - __len__ and __getitem__ methods
   - Lazy loading (load from disk per item, not all into memory)
   - Caching strategy for expensive preprocessing steps

2. DataLoader configuration:
   - num_workers: calculate optimal value based on CPU cores
   - pin_memory: True if using GPU
   - prefetch_factor: 2 (default) or higher if I/O bound
   - persistent_workers: True to avoid worker restart overhead
   - Appropriate batch size for available GPU memory

3. Data augmentation pipeline:
   - Training augmentations: {{augmentations}}
   - Validation augmentations: normalization only (no random augmentations)
   - Augmentations applied on CPU in workers, not on GPU

4. Memory efficiency:
   - Use memory-mapped files for large datasets if applicable
   - Stream from object storage (S3/GCS) without downloading fully if remote

5. Determinism:
   - Worker seed function to ensure reproducibility across runs

Return: complete Dataset and DataLoader code with comments explaining each design decision.