Domain-specific models are built or adapted to perform well on the vocabulary, document structures, and reasoning patterns of a particular field. Adaptation may occur through continued pretraining on domain corpora or through fine-tuning on task examples.
They typically trade general capability for improved performance within the domain.
Alternative Names:
Specialized Model, Vertical Model
Why it Matters?
The claim that a model is legally trained is often less meaningful than buyers assume, since retrieval quality and document processing usually affect accuracy more than domain adaptation of the base model. The question worth asking is what the adaptation actually improved and whether it was measured on documents resembling the buyer's. A model tuned on appellate opinions may perform no better on nursing home charts than a general model would.
Frequently Confused with
Related terms
Frequently asked questions
Does a legally trained model perform better on legal work?
What is the tradeoff with domain models?
