pytorch-architecture

What problem does it solve?

This Skill provides guidance and best practices for designing, implementing, and optimizing PyTorch neural network architectures, focusing on efficiency, memory usage, and training stability.

Core Features & Use Cases

• Architecture Sketching: Helps in defining tensor shapes and module structures for new models.
• PyTorch & Hardware Optimization: Offers advice on memory footprint estimation, vectorization, and GPU utilization.
• Training Script Standards: Enforces a strict initialization sequence for training scripts to ensure consistency and proper setup.
• Use Case: When developing a new transformer variant, use this Skill to ensure the forward pass is correctly sketched with shapes, memory usage is estimated for target hardware, and the training script adheres to the standard initialization sequence for distributed training and logging.

Quick Start

Use the pytorch-architecture skill to sketch the forward pass for a new encoder-decoder model, documenting tensor shapes at each step.