https://mlsys.wuklab.io/tags/tensor-parallelism/Recent content in Tensor Parallelism onHugo -- gohugo.ioenSat, 16 May 2026 00:00:00 +0000https://mlsys.wuklab.io/posts/nitsum/Sat, 16 May 2026 00:00:00 +0000https://mlsys.wuklab.io/posts/nitsum/A single LLM deployment now serves everything from latency-critical chat to relaxed background jobs under a fixed GPU budget. We designed <a href="https://arxiv.org/abs/2605.05467" target="_blank">Nitsum [arXiv &lsquo;26]</a>, the first serving system that treats <strong>tensor parallelism (TP) as a runtime control surface</strong> instead of a fixed deployment choice. By making TP switching nearly free and reconfiguring the cluster to track shifting workloads, Nitsum improves SLO-compliant <strong>goodput by up to 5.3x</strong> over state-of-the-art systems. <br/><br/> <a href="https://mlsys.wuklab.io/posts/nitsum/" target="_blank">Read More&hellip;</a>