OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Steerling

Steerling is an Apache-2.0-licensed causal diffusion language model project whose 8B model generates text and supports concept decomposition, attribution and steering.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Steerling — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Steerling when a project needs hands-on exploration of generation and concept-level analysis or steering, and its documented GPU and CUDA requirements fit the available environment.

Who it is for

  • Researchers examining how concepts contribute to language-model predictions.
  • Developers prototyping attribution or concept-based steering for diffusion language models.

Documented uses

  • Generate text with confidence-based iterative unmasking.
  • Decompose prediction contributions into known and discovered concepts.
  • Attribute generated tokens to input tokens or inspect attribution across chunks.
  • Intervene on concept activations to amplify or suppress features during generation.
  • Browse the model's concept vocabulary and extract representations.
Before you choose

Requirements & limits

The repository provides installation through PyPI or from source under the Apache-2.0 license. The documented quick start requires Python 3.13 or later, CUDA 12.8 and a GPU with at least 18 GB of VRAM.

  • The model is approximately 8B parameters and has a documented 4,096-token context length.
  • The quick-start configuration uses bfloat16 and CUDA; the stated minimum GPU memory is 18 GB.
  • The project provides separate base-model and instruction-tuned notebooks.
  • Concept decomposition distinguishes known concepts, discovered residual features and a reconstruction correction.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up Steerling notebooks

Install Steerling from its source repository with notebook dependencies for using the project’s analysis notebooks.

Before you start

  • Python 3.13 or later, as listed in the project quick-start requirements
  • The uv package manager

Steps

  1. Clone https://github.com/guidelabs/steerling.git and change into the steerling directory.

  2. Run the command below to install the notebook environment.

    Documented command
    uv sync --extra notebook

  3. Activate the environment with source .venv/bin/activate.

  4. Choose a documented notebook under notebooks/base_model/ or notebooks/instruct_model/ for generation, decomposition, attribution, or concept steering.

Expected outcome

The source checkout has the notebook dependency set installed, and the project’s documented notebooks are available to explore.

Guide limitations

  • The README identifies the notebook topics and locations but does not provide the contents or execution steps for each notebook in the installation instructions.
  • Running the base model’s generation example requires a CUDA GPU with at least 18 GB VRAM and CUDA 12.8.
  • The README says PyTorch is installed with CUDA 12.8 support by default; a different CUDA version requires manually installing PyTorch before installing Steerling.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
guidelabs/steerling · Official homepage title · retained in its source language
How we review sourcesSuggest a correction