Abstract:

We evaluate the capabilities of a selected large language model (LLM) in the named entity recognition task. With the recent advent of large language modeling and the creation of tools such as ChatGPT, it became possible to automate various language-processing tasks. Some LLMs are openly available and can be deployed locally, in an offline and privacy-oriented manner. Appraised model of the Llama 2 family is small enough to be efficiently run on a typical consumer machine. In the paper we present a workflow for performing NER tasks on LLMs, ascertain its suitability and summarize the means of improving it. This is done on preselected medical data.