Generative AI 2.0 Foundations, AGI, Applications, and Beyond

Our goal is to develop fundamental algorithmic and system framework for building and deploying orders of magnitude complex GenAI systems, compared to that of today, without creating a compute and energy crisis.

Cluster Members Anshumali Shrivastava (Lead PI) Richard Baraniuk Ankit Patel Collaborators Todd Treangen Anastasios Kyrillidis Santiago Segarra

SPIN-OFFs

• ThirdAI Corp. VC Funding Raised USD 7.5M
  
• xMAD Corp. VC Funding Raised USD 1M
  

Publications

2024

• KV Cache is 1 Bit Per Channel: Efficient Large Language Model Inference with Coupled Quantization. [To appear]
  Tianyi Zhang, Jonah Wonkyu Yi, Zhaozhuo Xu, Anshumali Shrivastava
  Neural Information Processing Systems (NeurIPS) 2024.
  
• NoMAD-Attention: Efficient LLM Inference on CPUs Through Multiply-add-free Attention. [To appear]
  Tianyi Zhang, Jonah Wonkyu Yi, Bowen Yao, Zhaozhuo Xu, Anshumali Shrivastava
  Neural Information Processing Systems (NeurIPS) 2023.
  
• Accelerating Inference with Fast and Expressive Sketch Structured Transform. [To appear]
  Aditya Desai, Kimia Saedi, Apoorv Walia, Jihyeong Lee, Keren Zhou, Anshumali Shrivastava
  Neural Information Processing Systems (NeurIPS) 2024.
  
• Soft Prompt Recovers Compressed LLMs, Transferably. [pdf coming soon]
  Zhaozhuo Xu, Zirui Liu, Beidi Chen, Shaochen Zhong, Yuxin Tang, Jue WANG, Kaixiong Zhou, Xia Hu, Anshumali Shrivastava
  International Conference on Machine Learning (ICML) 2024.
  
• In defense of parameter sharing for model-compression .
  Aditya Desai and Anshumali Shrivastava
  International Conference on Learning Representations (ICLR) 2024.