skills.rest
goupup-ai
goupup-aiCOMMUNITY
Explore all skills by this author →
GitHub

system-profile

Profile scripts, processes, GPUs, memory, and interconnects for compute bottlenecks.

2|Updated Aug 12, 2025
Search Tags:#performance profiling#bottleneck analysis#system optimization#memory profiling#llm serving#gpu profiling#ml training
One-click install
npx skills add https://github.com/goupup-ai/miccai25 --skill system-profile-goupup-ai
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: system-profile
Source: https://github.com/goupup-ai/miccai25/tree/main/ARIS/skills/system-profile
Command: npx skills add https://github.com/goupup-ai/miccai25 --skill system-profile-goupup-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Manually identifying performance bottlenecks in scripts, running processes, GPUs, memory, and system interconnects is time-consuming and prone to oversight, especially for complex compute workloads like machine learning training and LLM serving.

Core Features & Use Cases

  • Multi-target profiling: Supports profiling Python scripts, running processes (by PID or service name), GPUs, memory, interconnects, and full frameworks like vLLM serving.
  • Flexible measurement methods: Combines external industry-standard tools (cProfile, nvidia-smi, torch.profiler, nsys etc.) with custom code instrumentation for deep, targeted analysis.
  • Structured reporting: Generates clear breakdowns of CPU overhead, memory usage, communication latency, and GPU compute utilization, with prioritized actionable recommendations.
  • Use case: If your LLM serving pipeline has unexpectedly high latency, use this skill to profile NCCL communication overhead and GPU kernel utilization to pinpoint the root cause.

Quick Start

Use the system-profile skill to analyze the performance of your training script train.py and identify the top bottlenecks.

Frequently Asked Questions about system-profile

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

FAQPage Schema
How do I identify performance bottlenecks in machine learning training scripts?▼

To identify performance bottlenecks in machine learning training scripts, you can profile CPU overhead, memory usage, and GPU compute utilization using standard tools like cProfile and torch.profiler combined with custom code instrumentation.

What is the best way to profile GPU utilization for LLM serving pipelines?▼

Profiling GPU utilization for LLM serving pipelines involves measuring NCCL communication overhead and GPU kernel utilization using nvidia-smi and nsys to pinpoint the root causes of unexpectedly high latency.

Can I profile a running process by PID or service name for system resource analysis?▼

Yes, you can profile running processes by PID or service name for system resource analysis, enabling targeted measurement of live memory usage, CPU overhead, and communication latency across active compute-intensive workloads.

Does system profiling work with standard tools like nvidia-smi and nsys?▼

Yes, system profiling integrates with standard external industry tools including nvidia-smi, nsys, cProfile, and torch.profiler, combining their outputs with custom code instrumentation to generate structured bottleneck reports.

How do I analyze NCCL communication overhead causing high latency in vLLM serving?▼

To analyze NCCL communication overhead causing high latency in vLLM serving, you profile the framework to generate a structured breakdown of communication latency and GPU compute utilization alongside prioritized actionable recommendations.

What are the limitations of using standard profiling tools for full framework serving analysis?▼

Standard profiling tools alone may lack targeted measurement for specific compute, memory, and communication overhead, requiring custom code instrumentation to capture deep interconnect bottlenecks within complex compute workloads.

Related Skills

Explore More Software Engineering Skills→
tbvjvsladlatbvjvsladla
18

adversarial-benchmark

Validate vLLM serving instance performance against deterministic benchmarks and external references.

Community
Advanced
swaruplabswaruplab
94

get-available-resources

Detect CPU, GPU, memory, and disk resources and output a .claude_resources.json report.

Community
Intermediate
hve4638hve4638

system-profile

Profile Python scripts, processes, or hardware across CPU, memory, GPU, and interconnect.

Community
Advanced
LancerLabLancerLab

perf-nsight-compute-analysis

Analyze NVIDIA Nsight Compute reports to classify GPU kernel bottlenecks.

Official
Intermediate