QK-Normed MLA: QK normalization without full key caching

Query-key (QK) normalization stabilizes attention by controlling the scale of queries and keys before the dot product, but is not immediately compatible with Multi-head Latent Attention (MLA). MLA achieves efficient decoding by caching low-dimensional latent states instead of full keys, whereas post-projection QK RMSNorm appears to require the fully projected key for every cached token. We show this apparent incompatibility is an implementation artifact, not an architectural constraint. RMSNorm decomposes into a static affine weight and a dynamic scalar RMS statistic. The static key-side weight can be absorbed into the MLA query-side projection; the dynamic key statistic reduces to one inverse-RMS scalar per token and KV group. The resulting formulation is exactly equivalent to explicit post-projection QK RMSNorm in exact arithmetic and preserves MLA's latent decode path. In our 400M runs trained for up to 100B tokens, QK-Normed MLA achieves lower training loss and better downstream accuracy than QK clipping, while H800 decode benchmarks show less than 2% latency overhead up to 256k context. These results make QK normalization a practical stabilization option for MLA models without requiring full-key caching.

DriftGuard: Mitigating Asynchronous Data Drift in Federated Learning

In real-world Federated Learning (FL) deployments, data distributions on devices that participate in training evolve over time. This leads to asynchronous data drift, where different devices shift at different times and toward different distributions. Mitigating such drift is challenging: frequent retraining incurs high computational cost on resource-constrained devices, while infrequent retraining degrades performance on drifting devices. We propose DriftGuard, a federated continual learning framework that efficiently adapts to asynchronous data drift. DriftGuard adopts a Mixture-of-Experts (MoE) inspired architecture that separates shared parameters, which capture globally transferable knowledge, from local parameters that adapt to group-specific distributions. This design enables two complementary retraining strategies: (i) global retraining, which updates the shared parameters when system-wide drift is identified, and (ii) group retraining, which selectively updates local parameters for clusters of devices identified via MoE gating patterns, without sharing raw data. Experiments across multiple datasets and models show that DriftGuard matches or exceeds state-of-the-art accuracy while reducing total retraining cost by up to 83%. As a result, it achieves the highest accuracy per unit retraining cost, improving over the strongest baseline by up to 2.3x. DriftGuard is available for download from https://github.com/blessonvar/DriftGuard.

ORGEval: Graph-Theoretic Evaluation of LLMs in Optimization Modeling

Formulating optimization problems for industrial applications demands significant manual effort and domain expertise. While Large Language Models (LLMs) show promise in automating this process, evaluating their performance remains difficult due to the absence of robust metrics. Existing solver-based approaches often face inconsistency, infeasibility issues, and high computational costs. To address these issues, we propose ORGEval, a graph-theoretic evaluation framework for assessing LLMs' capabilities in formulating linear and mixed-integer linear programs. ORGEval represents optimization models as graphs, reducing equivalence detection to graph isomorphism testing. We identify and prove a sufficient condition, when the tested graphs are symmetric decomposable (SD), under which the Weisfeiler-Lehman (WL) test is guaranteed to correctly detect isomorphism. Building on this, ORGEval integrates a tailored variant of the WL-test with an SD detection algorithm to evaluate model equivalence. By focusing on structural equivalence rather than instance-level configurations, ORGEval is robust to numerical variations. Experimental results show that our method can successfully detect model equivalence and produce 100\% consistent results across random parameter configurations, while significantly outperforming solver-based methods in runtime, especially on difficult problems. Leveraging ORGEval, we construct the Bench4Opt dataset and benchmark state-of-the-art LLMs on optimization modeling. Our results reveal that although optimization modeling remains challenging for all LLMs, DeepSeek-V3 and Claude-Opus-4 achieve the highest accuracies under direct prompting, outperforming even leading reasoning models.