Introducing OpenBioLLM-Llama3-70B & 8B: Saama’s AI Research Lab Released the Most Openly Available Medical-Domain LLMs to Date!

Here at Saama, we are continuously innovating, researching, and developing AI models for healthcare.  Saama’s AI Research Lab is an in-house team of dedicated AI researchers and innovators, led by Malaikannan Sankarasubbu, VP of AI Research.

Throughout the past 3 years, a research team led by Ankit Pal, meticulously curated a diverse custom dataset, collaborating with medical experts to ensure the highest quality. The dataset spans over three thousand healthcare topics and 10+ medical subjects.

The (AI) models underwent a rigorous two-phase fine-tuning process using the LLama-3 70B & 8B models as the base and leveraging Direct Preference Optimization (DPO) for optimal performance. The models were able to outperform (using nine different benchmarks for measurement) industry giants such as: Openai’s GPT-4, Google’s Gemini, Google’s Med-PaLM-1, Med-PaLM-2 and opensource Meditron-70B in the biomedical domain, setting a new state-of-the-art for models of their size. 

Introducing OpenBioLLM-Llama3-70B & 8B: Saama’s AI Research Lab Released the Most Openly Available Medical-Domain LLMs to Date!
*Graph showing how OpenBio LLM-70B & OpenBioLLM-8B performed against industry giants

What’s next for Saama’s AI Innovation Labs? In the coming months, we’ll introduce expanded medical domain coverage, longer context windows, better benchmarks, and multimodal capabilities.

Introducing OpenBioLLM-Llama3-70B & 8B: Saama’s AI Research Lab Released the Most Openly Available Medical-Domain LLMs to Date!
*Graph created by Bart DeWitt, Digital Health and Regenerative AI Expert

Want to learn more about Saama’s AI capabilities, and how they can accelerate your clinical trials? Learn more today!

To learn more about OpenBioLLM- Llama3-70B & 8B, visit Analytics Vidhya.

Recommended Reading

Within clinical data management (CDM), there are numerous data quality, structure, volume, and collection challenges that make the process overly complex and difficult to oversee.