← 研踪 PaperTrailarXiv:2611.24395 · v1
RSIBench-Infra: Can Agents Optimize Their Own Training and Inference Infrastructure at Production Scale?
首次提交 2026-11-18 21:29
版本更新 2026-11-18 21:29
雷达观测 2026-11-19 15:04
摘要
Improving the efficiency of training and inference infrastructure is a practical component of recursive self-improvement. Existing benchmarks for agent-driven optimization largely focus on individual software layers such as kernel, leaving unclear whether agents can deliver end-to-end performance gains under production workloads. We introduce RSIBench-Infra, a benchmark of 20 training and inference tasks spanning diverse model families and modalities, with deployments of up to 32 accelerators on NVIDIA GPUs and Ascend NPUs. Agents start from runnable baselines derived from official recipes and can modify the software stack across engine, kernel, and instruction levels. An OCI image-layer submission protocol captures these changes for reproducible deployment, while independent evaluation measures end-to-end performance under task-specific quality constraints. Across all task results, 66.2% exceed validated speedup. These results demonstrate that current agents can improve production-scale training and inference performance, but identifying effective optimization strategies and consistently translating local improvements into end-to-end gains remain challenging.
作者备注:22 pages. Project page: https://rsibench-infra.github.io
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-18 21:29 · 最新观测版本
RSIBench-Infra: Can Agents Optimize Their Own Training and Inference Infrastructure at Production Scale?
Improving the efficiency of training and inference infrastructure is a practical component of recursive self-improvement. Existing benchmarks for agent-driven optimization largely focus on individual software layers such as kernel, leaving unclear whether agents can deliver end-to-end performance gains under production workloads. We introduce RSIBench-Infra, a benchmark of 20 training and inference tasks spanning diverse model families and modalities, with deployments of up to 32 accelerators on NVIDIA GPUs and Ascend NPUs. Agents start from runnable baselines derived from official recipes and can modify the software stack across engine, kernel, and instruction levels. An OCI image-layer submission protocol captures these changes for reproducible deployment, while independent evaluation measures end-to-end performance under task-specific quality constraints. Across all task results, 66.2% exceed validated speedup. These results demonstrate that current agents can improve production-scale training and inference performance, but identifying effective optimization strategies and consistently translating local improvements into end-to-end gains remain challenging.
查看此版本解读 →查看此版本原文 ↗