JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Runtime benchmark guide

Testing memory and speed for a Mac AI that makes choices

This guide shows how to test an AI tool that makes choices on an Apple computer. It explains how to measure the trade-off between using less computer memory and getting faster answers.

Original by afshinmEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“Laya MPS” by afshinm. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

This project provides a way to run a specific AI tool on an Apple computer. The software acts like Jev, meaning it chooses from options rather than writing out a full answer. People use it to sort customer service tickets or score security reports on their own machine.

The guide tests different settings to show how memory affects speed. On one computer, a minimal mode used 0.74 gigabytes of memory and took 171 milliseconds to answer. A faster mode used 2.11 gigabytes and took 32 milliseconds. These memory numbers only count the AI program, excluding the operating system.

This is useful for people who need to balance computer memory against fast response times. The tool is not a general AI assistant that can write emails or answer open-ended questions. A dashboard lets you check the exact version, device, and limits the software is currently using.

Key takeaways

  1. Test different memory settings with the same examples to see how they change response times.
  2. Clearly state whether memory measurements include the operating system or just the specific software program.
  3. Make it easy to check the exact software version and settings while the program is running.

The author tested an M5 Pro computer, reporting 0.74 gigabytes of memory at 171 milliseconds per answer in minimal mode, and 2.11 gigabytes at 32 milliseconds in reduced mode. Reviewers did not reproduce these results, which excluded network delays and used fixed test questions.

GitHub repository · Source reviewed

Read the original guide Opens the author’s site in a new tab.