perf-torch-sync-free

Official

Make PyTorch code truly async by removing syncs.

AuthorNVIDIA
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Identify and eliminate host-device synchronizations in PyTorch code. Detects sync points (.item(), .cpu(), boolean indexing, torch.tensor on CUDA), classifies false vs true dependencies, provides sync-free alternatives.

Core Features & Use Cases

  • Detects and classifies CPU-GPU synchronization points in PyTorch workloads.
  • Provides actionable, sync-free alternatives to common patterns like .item(), .cpu(), and tensor transfers.
  • Guides integration into existing codebases with a step-by-step workflow and verification.

Quick Start

Run through a PyTorch workload to detect and remove host-device syncs, then verify performance gains with sync-debug mode and profiling tools.

Dependency Matrix

Required Modules

None required

Components

scriptsreferences

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: perf-torch-sync-free
Download link: https://github.com/NVIDIA/skills/archive/main.zip#perf-torch-sync-free

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.