IBM Research, in partnership with Red Hat, has unveiled InstructLab, an innovative open-source initiative that enables collaborative customisation of large language models (LLMs) without requiring complete retraining. This endeavor seeks to expedite the incorporation of community contributions into basic models, greatly decreasing the time and effort previously necessary.

 

InstructLab works by combining human-curated data with high-quality examples generated by an LLM, decreasing the cost of data generation. This data can then be used to improve the underlying model without having to retrain it from scratch, resulting in significant cost savings. IBM Research has previously used InstructLab to produce synthetic data to improve its open-source Granite language and code models.

 

“There’s no good way to combine all of that innovation into a coherent whole,” said David Cox, vice president for AI models at IBM Research.

 

Researchers have used InstructLab to improve an IBM 20B Granite code model, transforming it into an expert in updating software created for IBM Z mainframes. This process exhibited speed and effectiveness, prompting IBM to develop a strategic collaboration with Red Hat.

 

IBM's current mainframe modernization tool, the Watsonx Code Assistant for Z, was fine-tuned using COBOL-Java programming. These were amplified using typical rules-based synthetic generators and further improved with InstructLab's capabilities.

 

“The most exciting part of InstructLab is its ability to generate new data from traditional knowledge sources,” noted Ruchir Puri, chief scientist at IBM Research. An updated version of WCA for Z is expected to be released soon.