steer·ability
The steerability toolkit is an open source Python package for steering large language models.
The toolkit enables the development and evaluation of a wide range of steering methods through an expressive library of reusable components across four model control surfaces (input, structure, state, and output). Features include the modular construction of novel steering methods, composition of steering methods into steering pipelines, and benchmarking of pipelines on custom use cases and metrics (including measurement of steering side effects). Check out our repo for more info.
News
-
Starting Aug 15!NeurIPS 2026 Competition: The Steerability Challenge
Design steering pipelines that reduce sycophancy while preserving general capabilities. Competition window August until October; in-person workshop to be held at NeurIPS 2026.
- Today · Aug 5, 2026
-
Jul 7, 2026 · San DiegoACL 2026 Systems Demo
A system demonstration of the toolkit: building, composing, and benchmarking steering methods across the four control surfaces.
-
Jan 21, 2026 · SingaporeAAAI 2026 Lab: Learning to Steer Large Language Models
A hands-on lab on inspecting model behavior, designing steering controls, and validating the steered behavior.