Tiny models, scored on your own machine

System 1 decision models answer typed questions with probabilities. System 2 models write answers, once with thinking off and once with it on. Every row was run locally with lvwb and is tied to the exact weights.

Add your model

1. Install

brew trust gitguy268/tap
brew install gitguy268/tap/lvw-bench

Or a release binary for macOS, Linux or Windows.

2. Run

lvwb run org/your-model

HF link or id, local .gguf, safetensors folder or a server URL.

3. Share

lvwb submit <result.json>

Prints the steps for a discussion on the dataset. Nothing is uploaded for you.