Does it show which models are currently taking up memory?

Yes, it uses the Ollama API's process status endpoint to list exactly which models are loaded and how many gigabytes of VRAM each one is consuming.

Will this help me diagnose why my model is slow?

Absolutely. The skill includes troubleshooting steps to check if a model is partially offloaded to the CPU and verifies if the GPU is actually reaching high utilization levels.

Can I monitor both NVIDIA and AMD GPUs?

Yes, the skill includes built-in support for both nvidia-smi and rocm-smi to provide status reporting across different hardware architectures.

How does the skill calculate inference speed?

It captures the wall-clock time and internal Ollama metadata, such as prompt evaluation count and generation duration, to calculate real-time tokens-per-second metrics.

Ollama GPU Monitoring

Name: Ollama GPU Monitoring
Author: atrawog

byatrawog

0•

分析与监控

Monitors GPU health, VRAM utilization, and inference performance metrics for Ollama-based AI models.

The GPU monitoring skill provides deep visibility into your hardware acceleration layer when running local AI models via Ollama. It enables developers to track real-time NVIDIA and AMD GPU status, monitor VRAM allocation for specific models, and benchmark inference efficiency through tokens-per-second metrics. Whether you are debugging slow responses, troubleshooting 'out of memory' errors, or optimizing resource usage in containerized environments, this skill offers the necessary visibility to maintain a high-performance local AI development workflow.

主要功能

01Real-time NVIDIA and AMD GPU status tracking (NVIDIA-SMI and ROCm-SMI)

02Inference performance benchmarking including tokens per second calculation

030 GitHub stars

04Granular VRAM monitoring for individual models loaded in memory

05Automated system health checks for GPU hardware and Ollama server status

06Diagnostic tools for identifying CPU fallback and hardware bottlenecks

使用场景

01Debugging performance bottlenecks during local LLM development

02Monitoring resource consumption when running large 70B+ parameter models

03Verifying GPU acceleration and driver compatibility in containerized environments

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add atrawog/overthink-plugins gpu

For use in Claude.ai and ChatGPT