debug-cuda-crash
CommunityDebug CUDA crashes with API-logged insights.
Authorsunxxuns
Version1.0.0
Installs0
System Documentation
What problem does it solve?
This Skill provides a guided approach to diagnosing CUDA crashes by enabling and analyzing FlashInfer's API logs, turning cryptic failures into actionable insights.
Core Features & Use Cases
- API-based debugging: Capture, log, and inspect API calls and tensor metadata when a CUDA crash occurs.
- Reproducible workflows: Step-by-step instructions to reproduce crashes in PyTorch CUDA workloads across single or multi-GPU setups.
- Targeted diagnostics: Quickly identify offending inputs, shapes, and steps leading to illegal memory accesses, NaN/Inf, or out-of-bounds errors.
Quick Start
- Enable basic logging: export FLASHINFER_LOGLEVEL=1; export FLASHINFER_LOGDEST=stdout
- Run a CUDA-enabled PyTorch script that exercises a FlashInfer API (e.g., batch_decode_with_padded_kv_cache) to reproduce the crash
- Open the logs to locate the API call trace and the tensor metadata that point to the root cause
Dependency Matrix
Required Modules
None requiredComponents
Standard package💻 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: debug-cuda-crash Download link: https://github.com/sunxxuns/flashinfer/archive/main.zip#debug-cuda-crash Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
Agent Skills Search Helper
Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.