aoti-debug
Diagnose and fix AOTInductor compilation and runtime errors in PyTorch, focusing on device and shape consistency.
Install
mkdir -p .claude/skills/aoti-debug && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/4024" && unzip -o skill.zip -d .claude/skills/aoti-debug && rm skill.zipInstalls to .claude/skills/aoti-debug
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Debug AOTInductor (AOTI) errors and crashes. Use when encountering AOTI segfaults, device mismatch errors, constant loading failures, or runtime errors from aot_compile, aot_load, aoti_compile_and_package, or aoti_load_package.Key capabilities
- →Validate device and shape matching for AOTI
- →Identify problematic kernels using intermediate value debuggers
- →Force deterministic CUDA behavior for debugging
- →Check input properties with runtime flags
How it works
It routes errors to specific debugging guides and provides environment variables to validate input guards, force synchronous execution, and print intermediate kernel values.
Inputs & outputs
When to use aoti-debug
- →Fix AOTI segfaults
- →Resolve device mismatch errors
- →Debug compile failures
About aoti-debug
Provides a structured approach to identifying AOTI failures. It enforces checks for device matching between compile/load and validates input shape requirements.
Debug AOTInductor (AOTI) errors and crashes. Use when encountering AOTI segfaults, device mismatch errors, constant loading failures, or runtime errors from aot_compile, aot_load, aoti_compile_and_package, or aoti_load_package.
When not to use it
- →When the error is unrelated to AOTInductor compilation or loading
- →When cross-device loading is required
Prerequisites
Limitations
- →Cross-device loading is not supported
- →CUDA illegal memory access errors can be non-deterministic
How it compares
It provides a systematic routing approach for AOTI-specific crashes rather than generic PyTorch debugging.
Compared to similar skills
aoti-debug side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| aoti-debug (this skill) | 1 | 9mo | Review | Advanced |
| langsmith-observability | 4 | 10mo | Review | Intermediate |
| mflux-debugging | 3 | 5mo | No flags | Advanced |
| debug-distributed | 1 | 9mo | Review | Advanced |
Try saying
Example prompts that trigger this skill in your AI assistant.
More by pytorch
View all by pytorch →You might also like
langsmith-observability
davila7
LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.
mflux-debugging
filipstrand
Debug MLX ports by comparing against a PyTorch/diffusers reference via exported tensors/images (export-then-compare).
debug-distributed
inclusionAI
Guide for debugging distributed training issues in AReaL. Use when user encounters hangs, wrong results, OOM, or communication errors.
benchmark-kernel
flashinfer-ai
Guide for benchmarking FlashInfer kernels with CUPTI timing
debug-model
vladmandic
Debug a broken SD.Next or Diffusers model integration. Use when a newly added or ported model fails to load, misdetects, crashes during prompt encoding or sampling, or produces incorrect outputs.
tokenizer-checker
openvinotoolkit
Validate a HuggingFace tokenizer with OpenVINO Tokenizers and OpenVINO GenAI. Use when: checking if a tokenizer converts and works correctly, verifying tokenizer/detokenizer accuracy, testing normalization steps, checking GenAI Tokenizer compatibility.