Bi-Daily Research Intelligence Briefing

Li et al. introduce MIPLIB-NL, a benchmark of 223 industrial-scale MILP instances (up to 10^7 variables) reverse-engineered from MIPLIB 2017, enforcing strict model-data separation. Results are sobering: SOTA models like GPT-4 and fine-tuned OR-LLMs drop from ~90% accuracy on existing toy benchmarks to ~18% here, failing primarily on structural consistency and index handling at scale. For us, the key takeaway is their "Loop-Based Structural Scaffold" taxonomy—a method to compress massive industrial formulations into compact LLM prompts via model-data separation. This is a mandatory read for our OR-Bench project, as it demonstrates that current evaluations are effectively measuring overfitting to toy problems rather than genuine modeling capability.

Kravatskiy et al. introduce ImprovEvolve, a framework that restricts the LLM to evolving `improve()` (local search) and `perturb()` (mutation) operators, which are then executed by a fixed basin-hopping algorithm. They achieve new state-of-the-art results on Hexagon Packing and the Second Autocorrelation Inequality, demonstrating that this modular approach generalizes to unseen problem sizes where monolithic AlphaEvolve solutions fail. The critical insight is that LLMs are poor at designing global search logic and tuning hyperparameters (LLM edits actively harmed performance), so we should isolate the LLM to generating local moves while keeping the meta-heuristic framework deterministic. We should immediately apply this 'operator-only' evolution strategy to our ALNS research for VRP.

AMPD introduces a disaggregated serving framework tailored for multi-round LLM agents, utilizing an offline ILP solver to optimize resource allocation (TP/DP configurations) and an online adaptive routing mechanism to handle incremental prefill tasks. The results are strong, showing 67-340% improvements in SLO attainment over vLLM and NVIDIA Dynamo by dynamically routing incremental prefill to decode workers when slack exists. For our 'GPUSched' project, the key takeaway is the specific ILP formulation (Eq. 5) for partitioning prefill/decode resources under global GPU constraints, and the insight that multi-agent workflows create a unique 'incremental prefill' bottleneck that standard disaggregation handles poorly.