PyTorch Patterns Guide
Guides the agent to write PyTorch training code with device-agnostic, reproducible, and efficient patterns.
- Repository
- affaan-m/ECC
- Stars
- 295,127
- Licence
- MIT
- Files
- 1 file
What it does
When installed, the agent follows a set of PyTorch coding conventions when you ask it to write or review deep learning code. It covers writing code that runs on either CPU or GPU, setting random seeds so experiments repeat, annotating tensor shapes, structuring model classes, writing training and validation loops with mixed precision, and configuring data loaders for speed. The file is a written reference of good and bad examples rather than a tool that runs anything itself.
Great for
- Engineers writing or reviewing PyTorch training scripts
- Researchers who want experiments that are reproducible across runs
- Students learning idiomatic PyTorch structure and training loop patterns
Skip it if
- Non-PyTorch frameworks such as TensorFlow or JAX, where these patterns do not apply
- Production-grade ML systems needing deployment, monitoring or security review, since the file only covers training and data code
- Non-programmers looking for a turnkey model, as the skill only shapes the agent's code suggestions
Try asking
- “Write a PyTorch image classifier with a clean nn.Module structure and a reproducible seed setup”
- “Review my training loop for missing model.eval() calls or inefficient gradient zeroing”
- “Set up a custom Dataset and DataLoader that loads images from a folder quickly”
- “Add mixed precision training with gradient clipping to my existing PyTorch script”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
pytorch-patterns.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip pytorch-patterns.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip pytorch-patterns.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Name your PyTorch version and hardware (CPU, single GPU, multi-GPU) so the agent picks compatible patterns, since some examples use torch.amp APIs that vary by version
- Treat the set_seed example as a starting point: full determinism can slow training and may not cover every library you use
- Ask the agent to explain any suggested change so you can verify shape comments and logic against your own model
What's inside
- SKILL.md
Redistributed unmodified from affaan-m/ECC under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.