Harvard Systems Open Data
Open traces and telemetry for research on LLM serving, coding agents, and computing infrastructure.
Dataset catalog
Dataset catalog
Production datasets from Harvard systems research.
LLM serving · request trace
One year of request-level production traffic from Chutes for LLM workload, caching, and load-balancing research.
Request-level timing, model and user identifiers, serving instances, token counts, latency, time-to-first-token, and prefix-cache reuse.
Coding agents · hashed sessions
Hashed session-level data from production coding-agent workloads.
The paper and GitHub artifact will be linked when the dataset is released.
FreeInference · infrastructure
GPU fleet and inference-engine telemetry from FreeInference serving.
The dataset is forthcoming. Current GPU, host, and serving-engine metrics are available on OpenInfra.
About the hub
Harvard Systems Open Data collects production datasets for research on LLM serving, coding agents, and computing infrastructure. Each entry links to its paper and GitHub artifact for documentation, code, and data access.