Harvard Releases Dataset of 6.12 Billion Real-World LLM Inference Requests

Harvard's MadSys lab, working with Chutes and FreeInference, released a full year of production LLM inference metadata covering 6.12 billion requests and about 91 million users, offering a rare open resource for studying real serving workloads.

2026-09-20 ~ 2026-09-22 · 3 related posts