Zijie Liu

Jianing Deng, Song Wang, Dongwei Wang et al.

Mixture-of-Experts Large Language Models (MoE-LLMs) achieve strong performance but incur substantial memory overhead due to massive expert parameters. Mixed-precision quantization mitigates this cost by allocating expert-wise bit-widths based on their importance, approaching the accuracy-memory Pareto frontier and enabling extreme low-bit quantization. However, existing methods rely on layer-wise importance estimation and overlook router shifts induced by quantization, resulting in suboptimal allocation and routing. In this work, we propose Global Expert-level Mixed-precision Quantization (GEMQ) to overcome these limitations via (1) a global linear-programming formulation that captures model-wide expert importance based on quantization error analysis, and (2) efficient router fine-tuning to adapt routing to quantized experts. These components are integrated into a progressive quantization framework that iteratively refines importance estimation and allocation. Experiments demonstrate that GEMQ significantly reduces memory and accelerates inference with minimal accuracy degradation. Source code is available at https://github.com/jndeng/GEMQ .

Jiarong Liang, Zhiheng Lyu, Zijie Liu et al.

Executable software engineering data is valuable for training SWE agents, but scaling it remains difficult for two reasons: only a small fraction of real repository changes yield verifiable, high-signal task instances, and naively building repository-specific environments quickly becomes the dominant systems cost. We present SWE-Next, an execution-grounded framework for scalable SWE task and trajectory collection. On the data side, SWE-Next mines real merged pull requests, executes candidate base/merged commit pairs, and retains only those that produce strict test improvements without regressions, yielding self-verifying instances. It also applies strict submission gating so that collected trajectories remain evidence-driven rather than speculative. On the systems side, SWE-Next introduces reusable repo-quarter profiles, which reuse the same environment across nearby commits in time while keeping each task run separate and reproducible. Using only 30 hours and 639GB of environment storage, SWE-Next processes 3,971 seed repositories and 102,582 candidate commit pairs mined from real merged PRs to construct a dataset of 2,308 self-verifying instances. Experiments show that SWE-Next improves downstream pass@1 with fewer or comparable training trajectories, indicating that its gains come not from a stronger trajectory generator, but from higher-signal execution-grounded supervision and more efficient data collection.

Zijie Liu, Jie Peng, Jinhao Duan et al.

Sparse Mixture-of-Experts (SMoE) architectures are increasingly used to scale large language models efficiently, delivering strong accuracy under fixed compute budgets. However, SMoE models often suffer from severe load imbalance across experts, where a small subset of experts receives most tokens while others are underutilized. Prior work has focused mainly on training-time solutions such as routing regularization or auxiliary losses, leaving inference-time behavior, which is critical for deployment, less explored. We present a systematic analysis of expert routing during inference and identify three findings: (i) load imbalance persists and worsens with larger batch sizes, (ii) selection frequency does not reliably reflect expert importance, and (iii) overall expert workload and importance can be estimated using a small calibration set. These insights motivate inference-time mechanisms that rebalance workloads without retraining or router modification. We propose Replicate-and-Quantize (R&Q), a training-free and near-lossless framework for dynamic workload rebalancing. In each layer, heavy-hitter experts are replicated to increase parallel capacity, while less critical experts and replicas are quantized to remain within the original memory budget. We also introduce a Load-Imbalance Score (LIS) to measure routing skew by comparing heavy-hitter load to an equal allocation baseline. Experiments across representative SMoE models and benchmarks show up to 1.4x reduction in imbalance with accuracy maintained within +/-0.6%, enabling more predictable and efficient inference.

Changsheng Gao, Zijie Liu, Li Li et al.

Like image coding in visual data transmission, feature coding is essential for the distributed deployment of large models by significantly reducing transmission and storage burden. However, prior studies have mostly targeted task- or model-specific scenarios, leaving the challenge of universal feature coding across diverse large models largely unexplored. In this paper, we present the first systematic study on universal feature coding for large models. The key challenge lies in the inherently diverse and distributionally incompatible nature of features extracted from different models. For example, features from DINOv2 exhibit highly peaky, concentrated distributions, while those from Stable Diffusion 3 (SD3) are more dispersed and uniform. This distributional heterogeneity severely hampers both compression efficiency and cross-model generalization. To address this, we propose a learned peaky-to-balanced distribution transformation, which reshapes highly skewed feature distributions into a common, balanced target space. This transformation is non-uniform, data-driven, and plug-and-play, enabling effective alignment of heterogeneous distributions without modifying downstream codecs. With this alignment, a universal codec trained on the balanced target distribution can effectively generalize to features from different models and tasks. We validate our approach on three representative large models (LLaMA3, DINOv2, and SD3) across multiple tasks and modalities. Extensive experiments show that our method achieves notable improvements in both compression efficiency and cross-model generalization over task-specific baselines. All source code has been made available at https://github.com/chansongoal/DT-UFC.

Rana Muhammad Shahroz Khan, Zijie Liu, Zhen Tan et al.

Reinforcement Learning (RL) and Supervised Fine-Tuning (SFT) are the two dominant paradigms for enhancing Large Language Model (LLM) performance on downstream tasks. While RL generally preserves broader model capabilities (retention) better than SFT, it comes with significant costs: complex reward engineering, instability, and expensive on-policy sampling. In contrast, SFT is efficient but brittle, often suffering from catastrophic forgetting due to $\textbf{Supervision Mismatch}$: the divergence between the model's evolving policy and static training labels. We address this trade-off with $\textbf{Trajectory-Mixed Supervision (TMS)}$, a reward-free framework that approximates the on-policy benefits of RL by creating a dynamic curriculum from the model's own historical checkpoints. TMS minimizes $\textit{Policy-Label Divergence (PLD)}$, preventing the mode collapse that drives forgetting in standard SFT. Experiments across reasoning (MATH, GSM8K) and instruction-following benchmarks demonstrate that TMS effectively shifts the accuracy--retention Pareto frontier. While RL remains the gold standard for retention, TMS significantly outperforms standard and iterative SFT, bridging the gap to RL without requiring reward models or verifiers. Mechanistic analysis confirms that PLD drift accurately predicts forgetting and that TMS successfully mitigates this drift.

Pedro F. C. de Carvalho, Zijie Liu, Md Umar Hashmi et al.

The incorporation of fairness into the distribution network (DN) planning and operation has become a key goal of recent studies. The cost of implementing fairness, denominated the price of fairness (PoF), covers the efficiency that is renounced for attaining social cohesion through fair outcomes. Locational disparity makes fairness schemes emerge to level the consumers playing field. However, fairness encompasses a range of notions. From egalitarian to merit-based criteria, various metrics are implemented as a tool for measuring equitable utility distribution. These have different mathematical complexities, from linear to non-linear programming cases, which affect their overall applicability. Hence, this study compiles the overarching fairness notions and metrics, reviewing how these affect stakeholders and the inherent mathematical optimisation in resource allocation problems. The aim is to support consistent and transparent planning and decision-making within DN operations.

Zijie Liu, Xinyu Zhao, Jie Peng et al.

Current medical AI systems often fail to replicate real-world clinical reasoning, as they are predominantly trained and evaluated on static text and question-answer tasks. These tuning methods and benchmarks overlook critical aspects like evidence-based reasoning and handling distracting information. To bridge this gap, we introduce a novel benchmark that simulates real-world diagnostic scenarios, integrating noise and difficulty levels aligned with USMLE standards. Moreover, we explore dialogue-based fine-tuning, which transforms static datasets into conversational formats to better capture iterative reasoning processes. Experiments show that dialogue-tuned models outperform traditional methods, with improvements of $9.64\%$ in multi-round reasoning scenarios and $6.18\%$ in accuracy in a noisy environment. Our findings highlight dialogue tuning as a promising approach for advancing clinically aligned and robust medical AI systems.

Kangqi Ni, Zhen Tan, Zijie Liu et al.

Reinforcement Learning with Verifiable Rewards (RLVR), primarily driven by the Group Relative Policy Optimization (GRPO) algorithm, is a leading approach for enhancing the reasoning abilities of Large Language Models (LLMs). Despite its wide adoption, GRPO's gains are often inconsistent; for instance, a model may show significant improvement in one reasoning domain, like mathematics, yet remain stagnant in another, such as medicine. This inconsistency raises a critical question: under what conditions does GRPO improve reasoning and generalize out-of-distribution (OOD)? We investigate this from a data distribution perspective. We first prove theoretically that GRPO is a conservative reweighting scheme, bounded by the base model's distribution and thus unable to discover completely novel solutions. We further validate this in carefully designed controlled studies by training transformers from scratch, evaluating generalization across reasoning depth, input length, token representation, and compositionality. Our results provide a principled explanation for GRPO's boundaries: OOD improvement emerges only when the target task aligns with the model's pretrained biases, while gains on in-distribution (ID) tasks diminish as performance saturates. This reframes GRPO not as a universal reasoning enhancer but as a tool that sharpens pretraining biases. Our findings motivate future development of algorithms that can expand a model's capabilities beyond its pretraining origin.

Baixiang Huang, Zhen Tan, Haoran Wang et al.

Agents based on Large Language Models (LLMs) have demonstrated strong capabilities across a wide range of tasks. However, deploying LLM-based agents in high-stakes domains comes with significant safety and ethical risks. Unethical behavior by these agents can directly result in serious real-world consequences, including physical harm and financial loss. To efficiently steer the ethical behavior of agents, we frame agent behavior steering as a model editing task, which we term Behavior Editing. Model editing is an emerging area of research that enables precise and efficient modifications to LLMs while preserving their overall capabilities. To systematically study and evaluate this approach, we introduce BehaviorBench, a multi-tier benchmark grounded in psychological moral theories. This benchmark supports both the evaluation and editing of agent behaviors across a variety of scenarios, with each tier introducing more complex and ambiguous scenarios. We first demonstrate that Behavior Editing can dynamically steer agents toward the target behavior within specific scenarios. Moreover, Behavior Editing enables not only scenario-specific local adjustments but also more extensive shifts in an agent's global moral alignment. We demonstrate that Behavior Editing can be used to promote ethical and benevolent behavior or, conversely, to induce harmful or malicious behavior. Through extensive evaluations of agents built on frontier LLMs, BehaviorBench validates the effectiveness of behavior editing across a wide range of models and scenarios. Our findings offer key insights into a new paradigm for steering agent behavior, highlighting both the promise and perils of Behavior Editing.

Yunfei Ling, Zijie Liu, Jun Du et al.

An accurate evolution model is crucial for effective control and in-depth study of fusion plasmas. Evolution methods based on physical models often encounter challenges such as insufficient robustness or excessive computational costs. Given the proven strong fitting capabilities of deep learning methods across various fields, including plasma research, this paper introduces a deep learning-based magnetic measurement evolution method named PaMMA-Net (Plasma Magnetic Measurements Incremental Accumulative Prediction Network). This network is capable of evolving magnetic measurements in tokamak discharge experiments over extended periods or, in conjunction with equilibrium reconstruction algorithms, evolving macroscopic parameters such as plasma shape. Leveraging a incremental prediction approach and data augmentation techniques tailored for magnetic measurements, PaMMA-Net achieves superior evolution results compared to existing studies. The tests conducted on real experimental data from EAST validate the high generalization capability of the proposed method.

Hailiang Zhao, Shuiguang Deng, Zijie Liu et al.

Multi-access Edge Computing (MEC) is booming as a promising paradigm to push the computation and communication resources from cloud to the network edge to provide services and to perform computations. With container technologies, mobile devices with small memory footprint can run composite microservice-based applications without time-consuming backbone. Service placement at the edge is of importance to put MEC from theory into practice. However, current state-of-the-art research does not sufficiently take the composite property of services into consideration. Besides, although Kubernetes has certain abilities to heal container failures, high availability cannot be ensured due to heterogeneity and variability of edge sites. To deal with these problems, we propose a distributed redundant placement framework SAA-RP and a GA-based Server Selection (GASS) algorithm for microservice-based applications with sequential combinatorial structure. We formulate a stochastic optimization problem with the uncertainty of microservice request considered, and then decide for each microservice, how it should be deployed and with how many instances as well as on which edge sites to place them. Benchmark policies are implemented in two scenarios, where redundancy is allowed and not, respectively. Numerical results based on a real-world dataset verify that GASS significantly outperforms all the benchmark policies.