nvitop/examples

nvitop Examples

Runnable reference scripts that exercise the public nvitop API. Each subfolder is fully self-contained: one runnable .py file, a README.md, and (if extra dependencies are required) a requirements.txt.

Index

Folder	What it shows	Extra deps
monitor-minimal/	Minimal plain-text GPU monitor using Device.all().	—
monitor-colored/	Colored monitor with per-process snapshots.	—
monitor-web/	Minimal stdlib HTTP(S) dashboard backed by collect_in_background with a rotating buffer.	—
take-snapshots/	Every form of take_snapshots — NVML, CUDA, single-device, processes off.	—
collector-tensorboard/	Log ResourceMetricCollector output to TensorBoard around a tiny PyTorch loop.	torch, tensorboard
collector-csv/	Append ResourceMetricCollector samples to a CSV file via pandas.	pandas
collector-background/	Run the collector on a daemon thread via collect_in_background.	—
select-devices-api/	Programmatic CUDA device selection (nvitop.select_devices), the API behind nvisel.	—
ml-framework-callbacks/	ML-framework callbacks (Keras, Lightning) and a TensorBoard helper built on nvitop.	varies per file

Running an Example

# Folders without a requirements file only need `nvitop` itself
python3 examples/<folder>/<script>.py

# Folders with a requirements file
pip install -r examples/<folder>/requirements.txt
python3 examples/<folder>/<script>.py

The ml-framework-callbacks/ folder ships per-framework requirements-<framework>.txt files so you can pull in only the framework you actually use.