Choosing the LLM Model
Today, there are many large language models to choose from, such as OpenAI’s ChatGPT, Google’s Gemini, and Meta’s open-source Llama model.
As my app is medically (eczema) focused, I wanted to use a large language model that can better and more accurately respond to eczema and other related medical questions. In my research, I learned about Google’s MedLM model which is particularly trained for healthcare scenarios. As ChatGPT and Llama are not medically focused, I assumed MedLM would better help my needs and decided to use it over other models.
MedLM is developed by Google and is only available through Google Cloud’s Vertex AI service, so I decided to use Google Cloud as my app’s backend cloud service provider. This required me to set up a new Google Cloud account and handle many other administrative and non-administrative setups to allow me to use Google Cloud’s Vertex AI service and its APIs.
Google, for some reason, has restricted MedLM model usage to allow-listed customers only. This required me to open a support request to Google Cloud, to allow me to the MedLM model. As I waited for my request to get approved, I continued with my Xema app development by using Google’s Vertex AI REST APIs for Google’s Gemini model. which was available to use without any restrictions.
After many hours put in, I was able to successfully integrate my app with Google Vertex REST APIs to use Google’s Gemini model, while I waited for Google to approve my request to use the MedLM model.
After a few weeks, Google ultimately disapproved my request, stating my app use case was against the model’s usage policy. While I was very disappointed with the rejection, I moved on and continued to use Google’s Gemini LLM model for my app, as I had it successfully integrated and working already. I feel good about this decision as it allows me to leverage my work, and continue developing an initial version of my app. I hope to refocus on using a more advanced, medically focused model at a later time.
Comments