Pre-alpha · more coming soon
Foothills Labs
A foundation model lab. The long-term goal is training models from the ground up; the starting point is deliberately narrower — build the evaluation we trust before we train anything we care about.
The mission
01 · The mission
Not written yet. It ships when it says exactly what we mean — nothing here is filler, so this section stays short until then.
Models
02 · The range
Releases will be named for mountains. Elevation tracks capability tier; continent tracks regional and language specialization. The range is surveyed — nothing is released yet.
More coming soon. The range →
Tools
03 · Shipped
Each released artifact gets its own repository, Apache-2.0.
regexbench and labloop are published
on PyPI; the leaderboard has no results yet.
regexbench
Evaluate generated regular expressions: semantic equivalence, correctness, and ReDoS safety.
labloop
Agent-driven experiment loop: propose a change, run it time-boxed, keep it only if the metric improves.
regexleaderboard
Runs regexbench across models and publishes the
numbers — scores, methodology, and a re-run command. No
results yet; the table ships empty until there are.
Principles
04 · How we work
Open weights
What we train, we release. Weights you can download, run, and check — a model you cannot inspect is a claim, not a result.
Reproducible research
Seeds, configs, and data provenance ship with every result. Anything we publish, you can re-run.
Reliable systems
The harness comes before the model, and the pipeline is built to be re-run, not rebuilt. Boring infrastructure is a feature.
Simple is not the enemy of powerful
A small model with a reproducible number beats a big claim with none. We reach for the plainest thing that works.
Contact
05 · Write us
Something here useful to you, or wrong? Open an issue, or write — we read everything.
info@foothills-labs.com