Unidax, an artificial intelligence (AI) electronic document company, announced today (18th) that it has unveiled a small language model (sLLM) used in its artificial intelligence solution 'EasyMy AI Doctor' on its huggling face for free.
The newly released small language model (sLLM) was developed by pre-training Meta's LLaMA 3.1 model based on the "ultra-large AI healthcare Q&A data" released on the AI hub of the National Intelligence Agency (NIA) and fine-tuning such as SFT and DPO.
The development of the small language model (sLLM) included Wikipedia, 200 million words of "healthcare Q&A data" released on AI hubs, 2 billion words of "super giant AI corpus with improved Korean performance," and about 5,000 words of medical and legal books.
However, the published data is designed to help understand the disease and does not guarantee an accurate diagnosis and treatment method of the disease, and it stipulates that the diagnosis of the exact symptoms should be consulted with a doctor.
Since the establishment of the AI business headquarters in September 2022, Unidax has been working on various AI projects in various industries in addition to electronic documents, and presents Unidax's own learning data construction and AI model development methodology.
In particular, it is participating in and carrying out the NIA's 'super large AI ecosystem construction project', completing and carrying out a number of medical projects such as the development of 'table data voice conversion AI service for the visually impaired' and 'establishment of artificial intelligence model data for pre-diagnosis of the elderly with dementia', and is also participating in autonomous driving development projects.
[Copyright holder (c) YTN Unauthorized reproduction, redistribution and use of AI data prohibited]