About this Event

Pretrained language models underpin several AI applications, but their high computational cost for training limits accessibility. Initiatives such as BLOOM and StarCoder aim to democratize access to pretrained models for collaborative community development. However, such existing models face challenges: limited multilingual capabilities, continual pretraining causing catastrophic forgetting, whereas pretraining from scratch is computationally expensive, and compliance with AI safety and development laws. In this talk, we will present Aurora-M, a 15B parameter multilingual open-source model trained on English, Finnish, Hindi, Japanese, Vietnamese, and code. Continually pretrained from StarCoderPlus on 435 billion additional tokens, Aurora-M surpasses 2 trillion tokens in total training token count. It is the first open-source multilingual model fine-tuned on human-reviewed safety instructions, thus aligning its development not only with conventional red-teaming considerations, but also with the specific concerns articulated in the Biden-Harris Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. Aurora-M is rigorously evaluated across various tasks and languages, demonstrating robustness against catastrophic forgetting and outperforming alternatives in multilingual settings, particularly in safety evaluations.

[huggingface blog] [paper]

Speakers

Diganta Misra is a UNIQUE Scholar Research MSc at Mila, Montreal supervised by Prof Irina Rish within the CERC AAI Lab and is a Visiting Researcher at the Human Sensing Lab at Carnegie Mellon University, Pittsburgh. His current research interests span topics of constrained learning - sparsity, mixture of experts, lifelong learning, generative models, learning dynamics and code generation. Diganta is also an incoming ELLIS PhD Fellow at the Max Planck Institute in Tübingen. Beyond research, he is an avid football, F1 and e-sport follower.