|
MUST-READ
M=8
P=8
I=9
RoboPhD optimizes LLM agent evolution under tight evaluation budgets by replacing traditional validation sets with an Elo-based tournament on training data and allowing agents to evolve their own diagnostic instrumentation. The results are backed by solid empirical comparisons, outperforming GEPA and Autoresearch on 3 out of 4 benchmarks (ARC-AGI, Text2SQL, DocFinQA) under a strict 1,500 evaluation budget. The single most useful takeaway we can steal is 'self-instrumenting agents'—seeding the initial agent with print() statements and letting the evolutionary process grow its own logging to provide richer Actionable Side Information (ASI) to the LLM optimizer. This paper matters immensely for AlgoEvo; we should immediately test dropping our validation splits in favor of Elo tracking and implement self-instrumenting diagnostics to improve our evolutionary signal without increasing API costs.
|
|
DISCUSS
2026-04-06 | University of Washington, NVIDIA, Rice University, University of Waterloo, Cisco Research, Independent Researcher | 2604.04335
M=6
P=7
I=6
GENSERVE optimizes the co-serving of text-to-image and text-to-video diffusion models on shared GPUs using a dynamic programming scheduler that jointly manages step-level preemption, dynamic batching, and elastic sequence parallelism. The results are empirically backed, demonstrating up to a 44% improvement in SLO attainment over baselines like SRTF on an 8-GPU cluster. The most useful takeaway is their two-stage DP formulation: they first generate a small set of anchored candidate actions (hold, resume, scale SP) per request, then run a knapsack DP to maximize global SLOs in under 2ms. While this targets diffusion models rather than LLMs, the OR formulation for elastic sequence parallelism and batching under strict latency SLOs is directly applicable to our GPUSched project.
|