Abstract--The Mixture of Experts (MoE) architecture has emerged as a key technique for scaling Large Language Models by activating only a subset of experts per query. Deploying MoE on consumer-grade edge hardware, however, is constrained by limited device memory, making dynamic expert offloading essential. Unlike prior work that treats offloading purely as a scheduling problem, we leverage expert importance to guide decisions, substituting low-importance active experts with functionally similar ones already cached in GPU memory, thereby preserving accuracy. As a result, this design reduces memory usage and data transfer, while largely eliminating PCIe overhead. In addition, we introduce a scheduling policy that maximizes the reuse ratio of GPU-cached experts, further boosting efficiency. Our extensive evaluations show that, compared with state-of-the-art approaches, our method achieves a 48% reduction in decoding latency and maintains an expert cache hit rate above 60%, all while preserving nearly lossless accuracy. MoE architectures offer a promising approach for deploying Large Language Models (LLMs) on edge devices, addressing an increasingly critical need [31], [30], [22]. Y et, edge servers are often limited in computational capacity and GPU memory, restricting full model deployment and rapid [32], [39]. Compared with dense models that compute all parameters for every input, MoE architectures mitigate these constraints by partitioning feed-forward layers into multiple experts [19], activating only a sparse subset per token. This design thus can drastically reduces computation overhead. However, GPU memory limitations introduce a new bottleneck: experts must frequently be offloaded to CPU memory and repeatedly loaded back to the GPU, resulting in substantial inference latency.

Instruction tuning is a pivotal technique for aligning large language models (LLMs) with human intentions, safety constraints, and domain-specific requirements. This survey provides a comprehensive overview of the full pipeline, encompassing (i) data collection methodologies, (ii) full-parameter and parameter-efficient fine-tuning strategies, and (iii) evaluation protocols. We categorized data construction into three major paradigms: expert annotation, distillation from larger models, and self-improvement mechanisms, each offering distinct trade-offs between quality, scalability, and resource cost. Fine-tuning techniques range from conventional supervised training to lightweight approaches, such as low-rank adaptation (LoRA) and prefix tuning, with a focus on computational efficiency and model reusability. We further examine the challenges of evaluating faithfulness, utility, and safety across multilingual and multimodal scenarios, highlighting the emergence of domain-specific benchmarks in healthcare, legal, and financial applications. Finally, we discuss promising directions for automated data generation, adaptive optimization, and robust evaluation frameworks, arguing that a closer integration of data, algorithms, and human feedback is essential for advancing instruction-tuned LLMs. This survey aims to serve as a practical reference for researchers and practitioners seeking to design LLMs that are both effective and reliably aligned with human intentions.

Knowledge gaps and hallucinations are persistent challenges for Large Language Models (LLMs), which generate unreliable responses when lacking the necessary information to fulfill user instructions. Existing approaches, such as Retrieval-Augmented Generation (RAG) and tool use, aim to address these issues by incorporating external knowledge. Yet, they rely on additional models or services, resulting in complex pipelines, potential error propagation, and often requiring the model to process a large number of tokens. In this paper, we present a scalable method that enables LLMs to access external knowledge without depending on retrievers or auxiliary models. Our approach uses constrained generation with a pre-built prefix-tree index. Triples from a Knowledge Graph are verbalized in textual facts, tokenized, and indexed in a prefix tree for efficient access. During inference, to acquire external knowledge, the LLM generates facts with constrained generation which allows only sequences of tokens that form an existing fact. We evaluate our proposal on Question Answering and show that it scales to large knowledge bases (800 million facts), adapts to domain-specific data, and achieves effective results. These gains come with minimal generation-time overhead. ReFactX code is available at https://github.com/rpo19/ReFactX.

To bridge the temporal granularity gap in energy network design and operation based on Energy System Models, resampling of time series is required. While conventional upsampling methods are computationally efficient, they often result in significant information loss or increased noise. Advanced models such as time series generation models, Super-Resolution models and imputation models show potential, but also face fundamental challenges. The goal of time series generative models is to learn the distribution of the original data to generate high-resolution series with similar statistical characteristics. This is not entirely consistent with the definition of upsampling. Time series Super-Resolution models or imputation models can degrade the accuracy of upsampling because the input low-resolution time series are sparse and may have insufficient context. Moreover, such models usually rely on supervised learning paradigms. This presents a fundamental application paradox: their training requires the high-resolution time series that is intrinsically absent in upsampling application scenarios. To address the mentioned upsampling issue, this paper introduces a new method utilizing Generative Adversarial Transformers (GATs), which can be trained without access to any ground-truth high-resolution data. Compared with conventional interpolation methods, the introduced method can reduce the root mean square error (RMSE) of upsampling tasks by 10%, and the accuracy of a model predictive control (MPC) application scenario is improved by 13%.

Transformers have become the cornerstone of modern large-scale language models, but their reliance on softmax attention poses a computational bottleneck at both training and inference. Recurrent models offer high efficiency, but compressing the full sequence into a fixed-size and holistic representation can suffer from memory degradation in long contexts and limit fine-grained retrieval. To address this, we propose RAT, an intermediate design that bridges the efficiency of RNNs and capacity of attention. RAT partitions the input into chunks, applies recurrence within each chunk for local dependencies, and softmax-based attention across chunks for long-range interactions. This design mitigates memory degradation and enables direct access to distant tokens, while retaining computational efficiency. Empirically, with a chunk size of 16, the RAT block achieves a 7$\times$ improvement in training speed for 100K sequence length and 9$times$ in generation at the 4K position, while maintaining similar performance compared to standard attention. We demonstrate this by training 1.3B parameter models from scratch and performing large-scale evaluations, including short- and long-context benchmarks, as well as supervised fine-tuning~(SFT). We further propose a hybrid architecture that interleaves RAT with local attention. By combining efficient long-range modeling with strong local interactions, this hybrid design not only improves inference speed and reduces cache memory usage, but also consistently enhances performance and shows the overall best results. Code is available at https://github.com/CLAIRE-Labo/RAT.

Many engineering and scientific fields have recently become interested in modeling terms in partial differential equations (PDEs) with neural networks, which requires solving the inverse problem of learning neural network terms from observed data in order to approximate missing or unresolved physics in the PDE model. The resulting neural-network PDE model, being a function of the neural network parameters, can be calibrated to the available ground truth data by optimizing over the PDE using gradient descent, where the gradient is evaluated in a computationally efficient manner by solving an adjoint PDE. These neural PDE models have emerged as an important research area in scientific machine learning. In this paper, we study the convergence of the adjoint gradient descent optimization method for training neural PDE models in the limit where both the number of hidden units and the training time tend to infinity. Specifically, for a general class of nonlinear parabolic PDEs with a neural network embedded in the source term, we prove convergence of the trained neural-network PDE solution to the target data (i.e., a global minimizer). The global convergence proof poses a unique mathematical challenge that is not encountered in finite-dimensional neural network convergence analyses due to (i) the neural network training dynamics involving a non-local neural network kernel operator in the infinite-width hidden layer limit where the kernel lacks a spectral gap for its eigenvalues and (ii) the nonlinearity of the limit PDE system, which leads to a non-convex optimization problem in the neural network function even in the infinite-width hidden layer limit (unlike in typical neural network training cases where the optimization problem becomes convex in the large neuron limit). The theoretical results are illustrated and empirically validated by numerical studies.

Automated Essay Scoring (AES) and Automatic Essay Feedback (AEF) systems aim to reduce the workload of human raters in educational assessment. However, most existing systems prioritize numerical scoring accuracy over feedback quality and are primarily evaluated on pre-secondary school level writing. This paper presents Multi-Agent Argumentation and Grammar Integrated Critiquer (MAGIC), a framework using five specialized agents to evaluate prompt adherence, persuasiveness, organization, vocabulary, and grammar for both holistic scoring and detailed feedback generation. To support evaluation at the college level, we collated a dataset of Graduate Record Examination (GRE) practice essays with expert-evaluated scores and feedback. MAGIC achieves substantial to near-perfect scoring agreement with humans on the GRE data, outperforming baseline LLM models while providing enhanced interpretability through its multi-agent approach. We also compare MAGIC's feedback generation capabilities against ground truth human feedback and baseline models, finding that MAGIC achieves strong feedback quality and naturalness.

The successful training of deep neural networks requires addressing challenges such as overfitting, numerical instabilities leading to divergence, and increasing variance in the residual stream. A common solution is to apply regularization and normalization techniques that usually require tuning additional hyperparameters. An alternative is to force all parameters and representations to lie on a hypersphere. This removes the need for regularization and increases convergence speed, but comes with additional costs. In this work, we propose a more holistic, approximate normalization via simple scalar multiplications motivated by the tight concentration of the norms of high-dimensional random vectors. Additionally, instead of applying strict normalization for the parameters, we constrain their norms. These modifications remove the need for weight decay and learning rate warm-up as well, but do not increase the total number of normalization layers. Our experiments with transformer architectures show up to 40% faster convergence compared to GPT models with QK normalization, with only 3% additional runtime cost. When deriving scaling laws, we found that our method enables training with larger batch sizes while preserving the favorable scaling characteristics of classic GPT architectures.

This paper introduces VeriThoughts, a novel dataset designed for reasoning-based Verilog code generation. We establish a new benchmark framework grounded in formal verification methods to evaluate the quality and correctness of generated hardware descriptions. Additionally, we present a suite of specialized small-scale models optimized specifically for Verilog generation. Our work addresses the growing need for automated hardware design tools that can produce verifiably correct implementations from high-level specifications, potentially accelerating the hardware development process while maintaining rigorous correctness guarantees. Our code and data are available at \href{https://github.com/wilyub/VeriThoughts}{this URL}.

--Federated Active Learning (F AL) seeks to reduce the burden of annotation under the realistic constraints of federated learning by leveraging Active Learning (AL). As F AL settings make it more expensive to obtain ground truth labels, F AL strategies that work well in low-budget regimes, where the amount of annotation is very limited, are needed. In this work, we investigate the effectiveness of TypiClust, a successful low-budget AL strategy, in low-budget F AL settings. Our empirical results show that TypiClust works well even in low-budget F AL settings contrasted with relatively low performances of other methods, although these settings present additional challenges, such as data heterogeneity, compared to AL. In addition, we show that F AL settings cause distribution shifts in terms of typicality, but TypiClust is not very vulnerable to the shifts. We also analyze the sensitivity of TypiClust to feature extraction methods, and it suggests a way to perform F AL even in limited data situations.