Skip to content

Repository files navigation

Local LLM N-vs-N Benchmark

ci license python


Serves GGUFs from ~/.lmstudio/models/ through a llama-swap proxy in front of llama.cpp podman containers. Benchmarks quality, latency, and cost (energy) across tasks × prompt_variants × models. Judge modes: pairwise_all tournament or scored rubric. Emits JSON, Markdown, and a single-file HTML dashboard under results/.

Quick start

./run.sh

Prerequisites, install, and configuration: HUMANS.md.

Highlights

  • llama-swap + llama.cpp Vulkan podman image; OpenAI-compatible client
  • pairwise_all tournament or scored 1–5 rubric; heuristic fallbacks
  • Quality, latency, and energy-based cost metrics
  • JSON, Markdown, and single-file HTML dashboard output
  • Any number of models; deterministic seeded synthetic dataset
  • ./run.sh fetch pulls missing GGUFs from Hugging Face

Documentation

Document Audience Contents
HUMANS.md Operators & developers Prerequisites, install, run, configure, troubleshoot, clean up
AGENTS.md LLMs & contributors Design invariants, hardware caveats, judge-mode selection, editing conventions
CONTRIBUTING.md Contributors PR checklist, commit style
config.yaml Reference Server, models, prompts, dataset, judge, cost, output (inline comments)

License

Licensed under the terms in LICENSE.

About

Local LLM N-vs-N benchmark harness — llama.cpp via llama-swap, quality/latency/cost scoring.

Topics

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages