I wanted a way to compare recent LLMs without digging through a different paper or config for each one. archdb puts their architectures and pretraining recipes in one table, with links back to the sources.
Install with uv:
uv tool install git+https://github.com/acforvs/archdb
archdbBrowse models and compare their architecture and training fields. Search fields with /, find
models with m, and press ? for the keys. The explorer works offline on macOS and Linux.
Update the tool and its bundled data with uv tool upgrade archdb.
There are also commands for filtering models and exporting results:
archdb models --where 'moe.sparsity >= 32' --sort params.active --desc
archdb show deepseek-v3 --group moeSee Terminal usage for filtering, source lookup, KV settings and JSON/CSV output.
The data is also available without the website:
- models.csv — one row per model.
- facts.csv — individual values with their sources and quotes.
- archdb.json — the complete dataset.
See Using the data for field definitions, other exports, and calculation details.
Requires Python 3.11+, Node.js 24+, and uv.
uv sync
uv run archdb build
npm --prefix web ci
npm --prefix web run build
uv run archdb serveOpen localhost:8000. See Development for checks and deployment commands.
To add a model or fix a value, see CONTRIBUTING.md. You can also open an issue.
Code: MIT. Data: CC0 1.0. Quoted material keeps its original license.