Systems for AI Lab Georgia Tech

Blog

2026

Agentic Workloads for Inference Evaluation
Agentic Workloads for Inference Evaluation

Why simple chat benchmarks are not enough for inference performance evaluation, and how to model agentic workloads with branching, prefix reuse, bursty timing, token heterogeneity, and reproducible synthetic sessions.

2021

Introducing CompOFA
Introducing CompOFA

Fast & Efficient Training of Once-For-All (OFA) models.