Generate reproducible deep-learning workload traces by mapping production GPU-cluster traces to profiled executable workloads.
-
Updated
Jun 16, 2026 - Python
Generate reproducible deep-learning workload traces by mapping production GPU-cluster traces to profiled executable workloads.
A learned wait-time score cannot rank a queue by anything except requested size — a negative result with zero counterexamples over 45,432 dispatch instants on the reference platform (45,268 and zero on Linux). Every number regenerates from one command.
Benchmark for portability of LLM-serving scheduler rankings across workloads, operating regions, metrics, and SLO definitions.
To associate your repository with the workload-traces topic, visit your repo's landing page and select "manage topics."