pytorch-patterns

原文涵盖的主题

• PyTorch Development Patterns
• When to Activate
• Core Principles
• 1. Device-Agnostic Code
• 2. Reproducibility First
• 3. Explicit Shape Management
• Model Architecture Patterns
• Clean nn.Module Structure

Idiomatic PyTorch patterns and best practices for building robust, efficient, and reproducible deep learning applications. Writing new PyTorch models or training scripts Reviewing deep learning code Debugging training loops or data pipelines Optimizing GPU memory usage or training speed Setting up reproducible experiments Always write code that works on both CPU and GPU without hardcoding devices. Set all random seeds for reproducible results. Always document and verify tensor shapes. | Idiom | Description | |-------|-------------| | model.train() / model.eval() | Always set mode before train/eval | | torch.nograd() | Disable gradients for inference | | optimizer.zerograd(settonone=True) | More efficient gradient clearing | | .to(device) | Device-agnostic tensor/model placement | | torch.amp.autocast | Mixed precision for 2x speed | | pinmemory=True | Faster CPU→GPU data transfer | | torch.compile | JIT compilation for speed (2.0+) | | weightsonly=True | Secure model loading | | torch.manualseed | Reproducible experiments | | gradientcheckpointing | Trade compute for memory | Remember: PyTorch code should be device-agnostic, reproducible, and memory-conscious. When in doubt, profil

这里只展示节选，请以完整原文和当前要求为准。