This guide helps you fix crashes and errors when using AOTInductor in PyTorch. You may see segfaults, device mismatch, or constant loading failures. The skill walks you through common problems and their solutions. It is built for developers who run into runtime errors after compiling models with aot_compile or aot_load.
You will learn to check that the compile device matches the load device. You will also check that input shapes and devices are correct. Following these steps helps you avoid many errors quickly. For trickier issues like CUDA illegal memory access, there are special flags to get clear error messages.
Use this skill whenever your AOTI code crashes or produces wrong output. It saves time by pointing you to the most likely causes first.
Global
mkdir -p ~/.claude/skills/aoti-debugProject
mkdir -p .claude/skills/aoti-debugSource Repository
Azure Diagnosticsmicrosoft/azure-skills
Debug Azure production issues with systematic diagnostics and AI-powered tools
Azure Messagingmicrosoft/azure-skills
Troubleshoot and fix Azure messaging SDK errors for Event Hubs and Service Bus
Use My Browserxixu-me/skills
Work inside your live browser session for real page inspection and debugging
Diagnosemattpocock/skills
A step-by-step debug loop for finding and fixing hard bugs
Diagnosing Bugsmattpocock/skills
Build a tight feedback loop to find any bug fast
Systematic Debuggingobra/superpowers
Find the root cause of every bug before attempting any fix to save time
Safe Debuglllllllama/rigorpilot-skills
Debug deep learning failures safely with conservative diagnosis and minimal patches
Firebase Crashlyticsfirebase/agent-skills
Set up crash reporting for your Android or iOS app with ease