AUG 14, 2026 · PREPRINT
Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training
arXiv
This is an unreviewed systems engineering paper demonstrating a runtime optimization technique with reported speedup metrics, but lacks clinical or scientific validation, ground-truth comparison, or peer review.
Reported
Speedup vs. serial colocation1.23×–1.30×
Speedup vs. disaggregation1.57×–2.24×
GPU cluster size32