Home

PUMA Wiki

PUMA is a local-first benchmarking platform for open-weight Large Language Models applied to ICT Project Management tasks — issue triage, story-point estimation, and pairwise prioritization. It runs entirely on your hardware via Ollama, needs no external API account, and produces reproducible results with a complete sustainability footprint for every run. This Wiki is the user-facing reference for installing, running, interpreting, and optionally publishing PUMA benchmarks.

What's inside this Wiki

• Installation — getting PUMA running on Linux, macOS, and Windows via Docker.
• Quick Start — your first benchmark in under five minutes.
• Architecture — the six-layer modular design and how data flows through it.
• Running Benchmarks — exhaustive flag reference for puma run, scenario catalog, prompting strategies.
• Models and Datasets — curated model catalog, hardware profiles, dataset sources.
• Metrics and Sustainability — the seven metric families and the carbon-footprint methodology.
• Publishing Results — submitting verified results to PUMA Community.
• FAQ — common questions and quick answers.

Quick links

• Main README — install, command tables, project layout.
• PUMA Community — the public hub where PUMA users share benchmark results.
• Latest release — release notes, tagged binaries, and changelog.

Get help

• For bug reports and feature requests, open an issue at pumacp/puma/issues.
• For general inquiries about the project, write to pumacapstoneproject@gmail.com.