ECE Researchers Receive NSF CSSI Framework Award to Enhance Computing Center User Support 

School of Engineering Department of Electrical and Computer Engineering Assistant Professors Zhao Zhang, Hang Liu, and Bo Yuan are the recipients of a collaborative Cyberinfrastructure for Sustained Scientific Innovation (CSSI) Framework award from the National Science Foundation (NSF).  

The team, led by PI Zhang, is taking the lead on the nearly $3 million, three-year project, “hpcGPT: Enhancing Computing Center User Support with HPC-enriched Generative AI.” 

The project, Zhang says, aims to build a ChatGPT-like service to enable automatic ticket answering for national computing centers.  

“This is a large collaborative project involving universities – Rutgers and Princeton – and national computing centers – NCSA, OSC, SDSC, TACC,” Zhang says. “It’s a unique opportunity to unite the high-performance computing (HPC) community and to work closely with AI experts to solve the critical problems these centers face.  

“While it’s a big personal achievement for me and will help me expand my knowledge base in HPC and AI,” he adds, “it is more of an opportunity for cross-community collaboration.” Liu agrees, stressing that the project has the potential to “change our community on how to use HPC infrastructure.” 

The project will, according to Zhang, bring the power of generative AI to hpcGPT – a question answering service for academic computing centers providing HPC platforms to STEM researchers. “hpcGPT will be able to enhance the user support service quality and efficiency, decrease response time, without increasing human efforts.”  

Zhang also predicts that “ultimately, domain scientists will benefit from the high-quality HPC service with enhanced research productivity, while major national computing centers will benefit from more efficient use of their supercomputing systems, and an increase in their support teams’ service capacity.” 

Collaboration is Key 

As PI, Zhang leads co-PIs Liu and Yuan in collaborating with a Princeton team on large language model (LLM) fine-tuning, and the integration of augmented generation (RAG) and chain-of-thoughts (CoT) to LLM. 

He is also overseeing everything in the project from startup and planning to technical, administrative, and fiscal components. “Hang leads the evaluation team and design metrics to evaluate the quality and performance of hpcGPT with human and AI feedback, while Bo leads the multi-modal aspects of the project,” he explains. 

Work on the project began in August, with Liu reporting that the team has already tuned a preliminary LLM for the project. Zhang expands, “For now, we’re designing the data cleaning and anonymization workflow, and evaluating a suite of candidate learning methods.” 

ECE students will benefit from the project, with Liu noting that, “We will train our students and broaden our vision on the practical impacts of large language models.” In addition, each member of the ECE team is supervising a student receiving support from grant funds.