Introducing ``Meditron'', an open source large-scale language model specializing in ``medical care''
Meditron , a medical specialized LLM suite based on Meta's large-scale language model (LLM) ``Llama 2'' and trained with medical books, has been released. Although formal adoption is not yet recommended as it has just been released, it has been reported that its abilities in the medical field exceed those of GPT-3.5.
[2311.16079] MEDITRON-70B: Scaling Medical Pretraining for Large Language Models
GitHub - epfLLM/meditron: Meditron is a suite of open-source medical Large Language Models (LLMs).
Large-scale language models are models that can process human words and output natural words, and are used in various services such as the chat service 'ChatGPT', as well as models specialized in specific fields such as finance and science. Models are also appearing one after another.
Although various efforts have been made to create models specific to the medical field, most of the resulting models have been closed source, such as GPT-4, or have been small in scale. The LLM suite 'Meditron' was created by Zemin Chen of the Swiss Federal Institute of Technology Lausanne and others who noticed this situation, and the LLM ' Meditron-7B ' and ' Meditron-70B ' were released at the same time as the announcement. Ta.
Both Meditron-7B and Meditron-70B are models based on Llama 2 that are trained on documents including
When we conducted benchmarks to measure performance in the medical field, Meditron-70B outperformed the LLM 'GPT-3.5' used in the free version of 'ChatGPT' and Google's medical LLM 'Med-PaLM', and outperformed GPT-4 and Med-PaLM. -It is said that it showed performance approaching that of PaLM-2. With the release of a new open source LLM with this performance, it is hoped that the medical knowledge and reasoning abilities of LLM will become widely used. Specifically, it is expected to be used to assist in medical examinations and diagnoses, and to query disease information.
At the time of writing, the Meditron suite is recommended for testing and evaluation purposes and is not suitable for use in production environments, either safely or professionally. It also adds that while the model can be used to generate text, it should not be used directly in the production of drugs or other tasks that may affect people.
Related Posts:
in Software, Posted by log1p_kr