skills.rest
llmh333
llmh333COMMUNITY
Explore all skills by this author →
GitHub

pytorch-patterns

Apply idiomatic PyTorch patterns for device-agnostic training and reproducible checkpoints.

Updated Sep 13, 2025
Search Tags:#checkpointing#pytorch#reproducibility#training loop#gpu memory#mixed precision#data loader
One-click install
npx skills add https://github.com/llmh333/employee_management_spring --skill pytorch-patterns-llmh333
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: pytorch-patterns
Source: https://github.com/llmh333/employee_management_spring/tree/main/.gemini/skills/pytorch-patterns
Command: npx skills add https://github.com/llmh333/employee_management_spring --skill pytorch-patterns-llmh333

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you avoid common PyTorch failures (non-reproducible results, device-specific bugs, shape mistakes, broken training/eval behavior, and fragile checkpointing) by standardizing correct patterns for real training pipelines.

Core Features & Use Cases

  • Device-agnostic development: Ensures your models and tensors move correctly across CPU/GPU without hardcoding .cuda() or device assumptions.
  • Reproducibility controls: Provides a complete seeding approach to make experiments repeatable.
  • Correct model & training structure: Guides clean nn.Module design, explicit tensor shape management, and safe train/eval switching.
  • Production-minded training loops: Covers mixed precision (AMP), gradient clipping, and efficient validation with torch.no_grad() and model.eval().
  • Efficient data pipelines & checkpointing: Shows custom Dataset/collate_fn patterns, optimized DataLoader settings, and resumable checkpoints with optimizer state.

Quick Start

Use the pytorch-patterns skill to review or rewrite a PyTorch training script so it uses device-agnostic code, reproducible seeding, correct train/eval modes, efficient data loading, and resumable checkpoints.

Frequently Asked Questions about pytorch-patterns

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make PyTorch training reproducible across different runs?▼

Make PyTorch training reproducible by applying comprehensive seeding across random number generators. This Skill provides patterns for complete seeding controls to ensure your experiments are repeatable and eliminate non-deterministic behavior.

What is the best way to handle device-agnostic code in PyTorch for CPU and GPU?▼

Device-agnostic PyTorch code uses `.to(device)` mappings without hardcoding `.cuda()`. This Skill provides patterns to ensure models and tensors move correctly across CPU and GPU environments, preventing device-specific training bugs.

How do I implement safe mixed precision and gradient clipping in a PyTorch training loop?▼

Implement safe mixed precision and gradient clipping in a PyTorch training loop using automatic mixed precision (AMP) patterns. This Skill guides production-minded loops with efficient validation via `torch.no_grad()` and `model.eval()`.

Why does my PyTorch model produce different results during validation versus training?▼

PyTorch models produce different results when `model.train()` and `model.eval()` modes are handled incorrectly. This Skill ensures proper train/eval switching and explicit tensor shape management to maintain consistent behavior.

How do I save and resume PyTorch checkpoints with optimizer state?▼

Save and resume PyTorch checkpoints with optimizer state by applying resumable checkpointing patterns. This Skill covers safe checkpointing practices to preserve optimizer state and seamlessly resume training pipelines.

Can I optimize PyTorch DataLoader settings for custom datasets and collate functions?▼

Optimize PyTorch DataLoader settings by using custom `Dataset` and `collate_fn` patterns. This Skill provides efficient data pipeline patterns to prevent bottlenecks and improve data loading performance.

Related Skills

Explore More Software Engineering Skills→
SnakeOSnakeO
9

refactor:pytorch

Refactor PyTorch code with modular nn.Module, torch.compile, AMP, and optimized DataLoader.

Community
Advanced
pluginagentmarketplacepluginagentmarketplace
5

deep-learning

Optimize neural network development and deployment workflows with PyTorch and TensorFlow.

Community
Advanced
vikrant-projectvikrant-project
2

weights-and-biases

Log machine learning experiments and models using the Weights & Biases API.

Community
Advanced
ovachieverovachiever
52

pytorch-fsdp

Apply PyTorch FSDP for sharded, mixed-precision distributed training of large models.

Community
Advanced