Teach Your AI New Tricks with InstructLab

 InstructLab is an open-source toolkit and workflow designed to fine-tune foundation models using a community-driven, taxonomy-based process.

 It enables experts—even those unfamiliar with ML—to contribute small “skill recipes” that enhance a model’s knowledge base.

How does it work?

1. Install the CLI tool(ilab)
1. pip install instructlab
2. Initialize a project
1. ilab config init  > It sets up a taxonomy repo and configuration.
3. Download a base model
1. ilab model download > such as IBM granite or Merlinite
2. It downloads pre-trained models from HuggingFace
4. List the downloaded model
1. ilab model list
5. Serve the model
1. ilab model serve
6. Chat with the model
1. ilab model chat

Contribute knowledge

1. Create skills via recipes 
1. Add examples(question and answer: YAML format), 
2. Add and validate your new data.
1. ilab taxonomy diff
3. Run synthetic data generation.
1. ilab data generate
4. Train or fine tune the model 
1. ilab model train --pipeline=simple
5. Serve the trained model
1. ilab lodel serve --model-path <path>
6. Chat with the model
1. ilab model chat --model <model>
7. Submit recipes as pull requests.