About this guide

Purpose and audience

This guide explains machine learning and model serving for technical readers without an ML background. It assumes comfort with numbers and simple formulas.

Numbers and sources

Measured examples come from local runs on an NVIDIA GB10. Some lack complete model, software, workload, or repeat-count details, so they are not portable performance targets. The GB10's specified memory bandwidth is 273 GB/s; a workload can achieve less.

Interactive figures use simplified models and illustrative values. Their notes describe the assumptions that matter for interpreting the results.

Limits

The guide emphasizes inference and serving. It does not cover data pipelines, distributed training, or reinforcement learning in depth.

Accessibility and privacy

The guide works offline and does not track readers. Search stays on the device. The theme follows the system setting, controls work with a keyboard, and motion respects the system preference. Figures include labels and text captions; some canvas details may not be accessible to all screen readers.

Corrections

To report an error, include the page, section, and proposed correction. For a measured result, include supporting evidence when available.