🥈 2nd place (Agent-only) MLSys 2026 Track B — Multi-Agent LLM Pipeline for Memory-Constrained Tensor DAG Scheduling. Neurosymbolic split: LLM for combinatorial structure (op fusion), deterministic solver for constraints ([w,h,k], dependency, latency). 7.77× aggregate speedup, 88× best case.
multi-agent important reflexion best-of-n dag-scheduling neurosymbolic llm-agents portfolio-search mlsys-2026 tensor-compilers
-
Updated
May 23, 2026 - Python