MBZUAI releases first LLM library

The Oryx Library contains three new domain-specific AI models

(Image credit: Carrington Malin via Midjourney)

MBZUAI (Mohamed bin Zayed University of Artificial Intelligence) has publicly released its first library of large-language models (LLMs) named the 𝗢𝗿𝘆𝘅 𝗟𝗶𝗯𝗿𝗮𝗿𝘆 via GitHub, to advance domain-specific applications. ClimateGPT, Video-ChatGPT and XrayGPT follow a number of LLM releases from Abu Dhabi-based R&D labs this year, including projects from Technology Innovation Institute (TII) and IIAI (Inception Institute of Artificial Intelligence)

The Oryx Library consists of projects and demos for large vision-language models developed at MBZUAI. The university's broad goal is to advance LLMs for multi-modal and domain-specific dialogues.

Three Oryx Library projects have just been released:

💻 𝗖𝗹𝗶𝗺𝗮𝘁𝗲𝗚𝗣𝗧 [English & Arabic] - a specialised LLM developed over the Vicuna framework for conversations related to Climate Change and Sustainability, in both English and Arabic languages. The goal is to develop an educational resource to support climate-change related conversations and help inform policymakers. ClimateGPT marks the first release of a large Arabic dataset (>500k samples) dedicated to climate change and sustainability. Codebase & demo available.

• Watch the ClimateGPT results video (Eng)

• Watch the ClimateGPT results video (Arabic)

💻 𝗩𝗶𝗱𝗲𝗼-𝗖𝗵𝗮𝘁𝗚𝗣𝗧 - a conversational engine for videos, based on a dedicated video-encoder and large language model (LLM) aligned using a simple linear projection, enabling video understanding and conversation about videos. The project includes the first set of high-quality 86k instruction data obtained specifically for video. Codebase, demo and a short results video available.

• Watch the Video-ChatGPT results video.

💻 𝗫𝗿𝗮𝘆𝗚𝗣𝗧 - a model for automated analysis of chest radiographs based on the given X-ray images. Frozen medical vision encoder (MedCLIP) is aligned with a fine-tuned LLM (Vicuna) using a simple linear transformation. The LLM is fine-tuned on medical data (100k real conversations between patients and doctors) and further on ~30k radiology conversations to acquire domain-specific features. XrayGPT generates interactive and clean summaries (~217k) from free-text radiology reports. Codebase, demo and a short results video available.

• Watch the XrayGPT results video.

The UAE has moved quickly to embrace the potential of LLMs. Earlier this month, the Artificial Intelligence, Digital Economy and Remote Work Applications Office published a guide for government workers using generative AI apps.

• Read the full MBZUAI blogpost:
  https://lnkd.in/dknzJAnC

• Check out the Oryx Library GitHub repositories:
  https://lnkd.in/daGnCgWQ

• Read my article on TII's Falcon LLM:
  https://lnkd.in/e7kzZNrj