At TechXchange 2024 IBM gave a nice preview how to run InstructLab in watsonx.ai. Below are some screenshots of a demo from the keynote.
InstructLab is an innovative methodology to customize smaller Large Language Models. To get started an open-source version is available. Read my blog post Customizations of Generative AI Models with InstructLab to understand more context.
Knowledge and Skills
In addition to fine-tuning InstructLab can be chosen to customize models.
New knowledge and skills can be defined in YAML files and submitted.
Review
Just like code extensions are sent to Git repositories via Pull Requests, new knowledge and skills can be sent to model owners to be merged into the baselines of models which can execute multiple AI tasks.
Synthetic Data
To fine-tune models with only little data, synthetic data is generated and evaluated.
Fine-tuning
Next the actual fine-tuning process is started.
Deployment
Fine-tuned models can be deployed on watsonx.ai so that they can be invoked by clients.
Next Steps
Check out the following resources to learn details.