Crafting Domain Specific Large Language Models with Enterprise Data

Advances in large language models have generated a tremendous amount of interest in their enterprise adoption. In many such situations, however, large language models must undergo significant customization and knowledge injection before they can be used in enterprise settings. Both business and technical decisions will be involved in the process of model customization, including the type of technology, the scale of training, as well as evaluation and annotation strategies. Unfortunately, many of these questions remain elusive to business decision-makers, as they must be addressed on a case-by-case basis. In this talk, we will discuss practical considerations in training domain-specific large language models, as well as their cost and benefit trade-offs. In particular, we will cover the life cycle of building a customized model, from the initial assessments, model training and tuning, evaluations and improvements, all the way to bug fixes and enhancements.