skills.rest
annealing-inversion
annealing-inversionCOMMUNITY
Explore all skills by this author →
GitHub

debug-cuda-crash

Enable Kernel API logging to diagnose CUDA crashes in SGLang.

Updated Jun 4, 2026
Search Tags:#debugging#logging#cuda#sglang#crash-diagnostics#kernel-api#tensor-metadata
One-click install
npx skills add https://github.com/annealing-inversion/sglang-kimi-deferral --skill debug-cuda-crash-annealing-inversion
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: debug-cuda-crash
Source: https://github.com/annealing-inversion/sglang-kimi-deferral/tree/main/.claude/skills/debug-cuda-crash
Command: npx skills add https://github.com/annealing-inversion/sglang-kimi-deferral --skill debug-cuda-crash-annealing-inversion

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

CUDA crashes in SGLang can be difficult to diagnose; this skill enables kernel API logging to capture inputs, metadata, and call boundaries that help identify root causes after a crash.

Core Features & Use Cases

  • Enable level-based Kernel API logging to see function boundaries, inputs, and outputs around CUDA calls.
  • Inspect tensor shapes, dtypes, devices, and statistics to verify data integrity before crashes.
  • Reproduce and diagnose issues in LLM and diffusion pipelines with crash-safe dumps and structured metadata.

Quick Start

Enable kernel API logging at level 3 and run your script to capture tensor metadata around the crash.

Frequently Asked Questions about debug-cuda-crash

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

FAQPage Schema
How do I debug CUDA crashes in SGLang when running LLM or diffusion pipelines?▼

Debug CUDA crashes in SGLang by enabling Kernel API logging to capture function boundaries, inputs, and tensor metadata around the crash. Level-based logging helps reproduce and diagnose issues in LLM and diffusion pipelines.

What log levels are available for SGLang kernel API crash diagnostics?▼

SGLang kernel API crash diagnostics support multiple log levels, such as level 3, to capture varying detail of tensor metadata and function boundaries. Higher levels provide deeper visibility into inputs, outputs, and call boundaries around CUDA calls.

How can I inspect tensor metadata to verify data integrity before a CUDA crash?▼

Inspect tensor metadata by enabling kernel API logging to capture shapes, dtypes, devices, and statistics before the crash. This verifies data integrity and helps identify invalid inputs or corrupted tensors around CUDA calls.

Does enabling kernel API logging work with both LLM and diffusion workflows in SGLang?▼

Kernel API logging works with both LLM and diffusion workflows in SGLang. It provides crash-safe dumps and structured metadata to reproduce and diagnose issues across different pipeline contexts.

How do I set up controlled crash dumps for CUDA kernel debugging in SGLang?▼

Set up controlled crash dumps by configuring dump directories and enabling kernel API logging in SGLang. This captures inputs, outputs, and tensor metadata safely during CUDA crashes without losing diagnostic data.

Related Skills

Explore More Software Engineering Skills→
OpenMOSSOpenMOSS
410

debug-cuda-crash

Log CUDA kernel API calls and tensor metadata around crash boundaries in SGLang.

Official
Intermediate
QingHuan-6QingHuan-6

debug-cuda-crash

Diagnose CUDA crashes in SGLang by exposing kernel API call traces and tensor metadata.

Community
Intermediate
hey-konghey-kong

debug-cuda-crash

Diagnoses CUDA crashes in SGLang via level-based kernel API logging and crash dumps.

Community
Advanced
dyyounggdyyoungg

debug-cuda-crash

Capture CUDA crash boundaries and tensor metadata via kernel API logging in SGLang.

Community
Advanced