InstructLab

An new community-based approach to build truly open-source LLMs

Weekly release cycle for models and data

Each week, the InstructLab model will be updated with the latest contributions of skills and knowledge from the community, resulting in a new state of the art model.

Each latest version will be shared on Hugging Face, and all synthetic data generated by the project to date will also be made available.

InstructLab
Experiment with adding skills on a local model
Submit perfected skills to InstructLab project
InstructLab project updates main model
New skills recipes
Approved skills
New SoTA model version

How it works

detect-gap doesn't exist
1
Experiment locally, on a quantized version of the latest InstructLab model, with adding new knowledge and skills
taxonomy doesn't exist
2
Once a synthetic data recipe is ready, submit a pull request to contribute back to the InstructLab repository, which is organized in a structured taxonomy
synthetic-data doesn't exist
3
InstructLab uses the new recipes to systematically generate synthetic data for the new capabilities - and this data is released to the community
fine-tune doesn't exist
4
InstructLab merges the new skill back into the base model by updating it using all synthetic data generated to date, and releases a new model version to the community

Join the community

Downloading the Lab CLI and experimenting with adding new skills and knowledge is a great way to get started. You can also get involved by contributing new skills, knowledge, or synthetic data directly.