AI Paper Atlas

Research map / Large language models / Agents, code and math

LEGO: An LLM-Enabled Hierarchical Optimizer for Tensor Computation Graphs with Structure-Aware Search and Compositional Synthesis

Ruiyuan Xu, Shuoming Zhang, Guangli Li, Qiuchu Yu, Rui Zhang, Yangyu Zhang, Hao Qian, Chunwei Xia, Jiacheng Zhao, Chenxi Wang, Xiaobing Feng, Jingling Xue and 1 more

ICML 2026 · Large language models › Agents, code and math · approach: LLM-guided program synthesis

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add bottom-up speculative fusion (mutation) to top-down construction (ablation)LEGO · mHC-MoE, H100relative speedup: 1.20× -> 11.29×Improved
add bottom-up speculative fusion (ablation)LEGO · SD3-MMDiT, H100relative speedup: 0.85× -> 2.86×Improved
allow deep recursive decomposition instead of depth-1 (ablation)LEGO · Mamba2, SD3-MMDiT, Qwen3-8B, H100codegen successImproved
allow deep recursive decomposition instead of depth-1 (ablation)LEGO · ResNet18, H100relative speedup: 0.93× -> 1.02×No clear effect
use structural decomposition instead of direct monolithic codegen (ablation)LEGO · 5 workloads, H100codegen successImproved
use LEGO instead of other LLM kernel agentsLEGO · Table 1speedupImproved
use LEGO instead of TorchInductor max-autotuneLEGO · mHC-MoE, H100speedup vs Eager: 0.82× -> 9.26×Improved
use LEGO instead of Inductor max-autotune on ResNet18LEGO · ResNet18, H100speedup vs Eager: 3.87× -> 3.96×No clear effect
use structure-aware FusionTree with context isolation instead of monolithic baselineLEGO · Mamba2-1.3Btoken consumption: 7.02× reductionImproved
parallel critical-path exploration instead of serial monolithic searchLEGO · 5 workloadsexploration time: 1.94× to 2.47× fasterImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy