Zehua Pei

Zehua Pei, Hui-Ling Zhen, Yu Zhang et al.

Large language models (LLMs) have improved Verilog generation from natural-language specifications, but most pipelines still treat generation as isolated sampling followed by functional checking. This is insufficient for practical RTL design, where useful Verilog must be correct, synthesizable, timing-conscious, and friendly to downstream hardware objectives. We present Verilog-Evolve, a feedback-driven framework for versioned Verilog refinement and cross-session skill evolution. For each task, Verilog-Evolve generates diverse minor candidates, evaluates them with executable feedback from functional simulation, Yosys synthesis, ABC timing proxy, and optional GEMM metrics, then promotes the best candidate into a major version under configurable scoring. To improve across tasks, the system maintains modular skill guidance, retrieves skills according to task and feedback context, and evolves candidate skills from logged histories through create/improve/skip decisions and verifier reports. Experiments on VerilogEval and mixed-precision GEMM tasks show that Verilog-Evolve improves final functional success and promotion stability while producing more downstream-friendly RTL under open-source synthesis, timing-proxy, and netlist-level GEMM objectives. Validation-gated skill evolution further improves GEMM downstream quality and achieves the best downstream score and GEMM held-out pass rate among the evaluated skill modes.

Zehua Pei, Hui-Ling Zhen, Shixiong Kai et al.

Large Language Model (LLM) agents are increasingly deployed in environments that generate massive, dynamic contexts. However, a critical bottleneck remains: while agents have access to this context, their static prompts lack the mechanisms to manage it effectively, leading to recurring Corrective and Enhancement failures. To address this capability gap, we introduce \textbf{SCOPE} (Self-evolving Context Optimization via Prompt Evolution). SCOPE frames context management as an \textit{online optimization} problem, synthesizing guidelines from execution traces to automatically evolve the agent's prompt. We propose a Dual-Stream mechanism that balances tactical specificity (resolving immediate errors) with strategic generality (evolving long-term principles). Furthermore, we introduce Perspective-Driven Exploration to maximize strategy coverage, increasing the likelihood that the agent has the correct strategy for any given task. Experiments on the HLE benchmark show that SCOPE improves task success rates from 14.23\% to 38.64\% without human intervention. We make our code publicly available at https://github.com/JarvisPei/SCOPE.

Zehua Pei, Hui-Ling Zhen, Weizhe Lin et al.

Diffusion Language Models (DLMs) offer attractive advantages over Auto-Regressive (AR) models, such as full-attention parallel decoding and flexible generation. However, they suffer from a notable train-inference mismatch: DLMs are trained with a static, single-step masked prediction objective, but deployed through a multi-step progressive denoising trajectory. We propose MemDLM (Memory-Enhanced DLM), which narrows this gap by embedding a simulated denoising process into training via Bi-level Optimization. An inner loop updates a set of fast weights, forming a Parametric Memory that captures the local trajectory experience of each sample, while an outer loop updates the base model conditioned on this memory. By offloading memorization pressure from token representations to parameters, MemDLM yields faster convergence and lower training loss. Moreover, the inner loop can be re-enabled at inference time as an adaptation step, yielding additional gains on long-context understanding. We find that, when activated at inference time, this Parametric Memory acts as an emergent in-weight retrieval mechanism, helping MemDLM further reduce token-level attention bottlenecks on challenging Needle-in-a-Haystack retrieval tasks. Code: https://github.com/JarvisPei/MemDLM.

Guojin Chen, Zehua Pei, Haoyu Yang et al.

Lithography is fundamental to integrated circuit fabrication, necessitating large computation overhead. The advancement of machine learning (ML)-based lithography models alleviates the trade-offs between manufacturing process expense and capability. However, all previous methods regard the lithography system as an image-to-image black box mapping, utilizing network parameters to learn by rote mappings from massive mask-to-aerial or mask-to-resist image pairs, resulting in poor generalization capability. In this paper, we propose a new ML-based paradigm disassembling the rigorous lithographic model into non-parametric mask operations and learned optical kernels containing determinant source, pupil, and lithography information. By optimizing complex-valued neural fields to perform optical kernel regression from coordinates, our method can accurately restore lithography system using a small-scale training dataset with fewer parameters, demonstrating superior generalization capability as well. Experiments show that our framework can use 31% of parameters while achieving 69$\times$ smaller mean squared error with 1.3$\times$ higher throughput than the state-of-the-art.

Zehua Pei, Hui-Ling Zhen, Xianzhi Yu et al.

Large language models can now process increasingly long inputs, yet their ability to effectively use information spread across long contexts remains limited. We trace this gap to how attention budget is spent during supervised fine-tuning (SFT) on long sequences: positional biases and attention sinks cause the model to allocate most of its attention to positionally privileged tokens rather than semantically relevant content. This training-time attention dilution (the starvation of content tokens in the attention distribution) weakens the gradient signal, limiting the model's ability to learn robust long-context capabilities. We introduce FocuSFT, a bilevel optimization framework that addresses this problem at training time. An inner loop adapts lightweight fast-weight parameters on the training context to form a parametric memory that concentrates attention on relevant content, and the outer loop performs SFT conditioned on this sharpened representation. Both loops apply bidirectional attention over context tokens while preserving causal masking for responses, reducing the causal asymmetry that gives rise to attention sinks and aligning inner-outer behavior. On BABILong, FocuSFT improves accuracy by up to +14pp across 4K--32K context lengths; on RULER, it raises CWE aggregation from 72.9\% to 81.1\% at 16K; and on GPQA with agentic tool use, it yields a 24\% relative gain in pass@1. Attention analysis shows that FocuSFT reduces attention sink mass by 529$\times$ and triples context engagement during training. Code: https://github.com/JarvisPei/FocuSFT

Zhentao Tang, Yuqi Cui, Shixiong Kai et al.

Expert-level scientific reasoning remains challenging for large language models, particularly on benchmarks such as Humanity's Last Exam (HLE), where rigid tool pipelines, brittle multi-agent coordination, and inefficient test-time scaling often limit performance. We introduce ReThinker, a confidence-aware agentic framework that orchestrates retrieval, tool use, and multi-agent reasoning through a stage-wise Solver-Critic-Selector architecture. Rather than following a fixed pipeline, ReThinker dynamically allocates computation based on model confidence, enabling adaptive tool invocation, guided multi-dimensional reflection, and robust confidence-weighted selection. To support scalable training without human annotation, we further propose a reverse data synthesis pipeline and an adaptive trajectory recycling strategy that transform successful reasoning traces into high-quality supervision. Experiments on HLE, GAIA, and XBench demonstrate that ReThinker consistently outperforms state-of-the-art foundation models with tools and existing deep research systems, achieving state-of-the-art results on expert-level reasoning tasks.

Zehua Pei, Hui-Ling Zhen, Ying Zhang et al.

Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them. This paper introduces a novel ``Behavioral Fingerprinting'' framework designed to move beyond traditional evaluation by creating a multi-faceted profile of a model's intrinsic cognitive and interactive styles. Using a curated \textit{Diagnostic Prompt Suite} and an innovative, automated evaluation pipeline where a powerful LLM acts as an impartial judge, we analyze eighteen models across capability tiers. Our results reveal a critical divergence in the LLM landscape: while core capabilities like abstract and causal reasoning are converging among top models, alignment-related behaviors such as sycophancy and semantic robustness vary dramatically. We further document a cross-model default persona clustering (ISTJ/ESTJ) that likely reflects common alignment incentives. Taken together, this suggests that a model's interactive nature is not an emergent property of its scale or reasoning power, but a direct consequence of specific, and highly variable, developer alignment strategies. Our framework provides a reproducible and scalable methodology for uncovering these deep behavioral differences. Project: https://github.com/JarvisPei/Behavioral-Fingerprinting

Zehua Pei, Lancheng Zou, Hui-Ling Zhen et al.

Scaling large language models (LLMs) improves performance but dramatically increases inference costs. The feed-forward network (FFN), consuming approximately 70\% of inference compute, represents a critical bottleneck, particularly in large batch size scenarios. While mixture-of-experts (MoE) architectures leverage activation sparsity for efficiency, converting existing dense models to MoEs traditionally requires resource-intensive continual pre-training. We present CMoE, a framework that rapidly transforms dense LLMs into MoEs without training. The key innovation lies in analyzing FFN neuron activations to partition them into shared (always active) and routed experts. Routed neurons are clustered using a balanced assignment algorithm, and a differentiable router is constructed analytically from activation statistics, enabling immediate deployment or optional lightweight fine-tuning. Experiments demonstrate that, with activation ratio of 75\%, it achieves remarkable results, delivering lossless precision in terms of perplexity while still maintaining a 5\% acceleration. Further experiments reveal that a CMoE configuration activating just 25\% of parameters reduces end-to-end latency by 1.5x while preserving usable perplexity without additional training. Moreover, a brief LoRA fine-tuning process (requiring only 1 hour and 2,000 samples) successfully recovers over 76\% of the dense model's downstream accuracy. By effectively balancing performance and efficiency, CMoE offers a viable path forward for deploying LLMs in real-world scenarios where computational resources are limited. We make our code publicly available at https://github.com/JarvisPei/CMoE.

Zehua Pei, Hui-Ling Zhen, Mingxuan Yuan et al.

Due to the growing complexity of modern Integrated Circuits (ICs), there is a need for automated circuit design methods. Recent years have seen rising research in hardware design language generation to facilitate the design process. In this work, we propose a Verilog generation framework, BetterV, which fine-tunes the large language models (LLMs) on processed domain-specific datasets and incorporates generative discriminators for guidance on particular design demands. The Verilog modules are collected, filtered and processed from internet to form a clean and abundant dataset. Instruct-tuning methods are specially designed to fine-tune the LLMs to understand the knowledge about Verilog. Furthermore, data are augmented to enrich the training set and also used to train a generative discriminator on particular downstream task, which leads a guidance for the LLMs to optimize the Verilog implementation. BetterV has the ability to generate syntactically and functionally correct Verilog, which can outperform GPT-4 on the VerilogEval benchmark. With the help of task-specific generative discriminator, BetterV can achieve remarkable improvement on various electronic design automation (EDA) downstream tasks, including the netlist node reduction for synthesis and verification runtime reduction with Boolean Satisfiability (SAT) solving.

Lancheng Zou, Shuo Yin, Zehua Pei et al.

Channel permutation is a powerful technique for enhancing the accuracy of N:M sparse models by reordering the channels of weight matrices to prioritize the retention of important weights. However, traditional channel permutation methods rely on handcrafted quality metrics, which often fail to accurately capture the true impact of pruning on model performance. To address this limitation, we propose PermLLM, a novel post-training pruning framework that introduces learnable channel permutation (LCP) for N:M sparsity. LCP leverages Sinkhorn normalization to transform discrete permutation matrices into differentiable soft permutation matrices, enabling end-to-end optimization. Additionally, PermLLM incorporates an efficient block-wise channel permutation strategy, which significantly reduces the number of learnable parameters and computational complexity. PermLLM seamlessly integrates with existing one-shot pruning methods to adaptively optimize channel permutations, effectively mitigating pruning-induced errors. Extensive experiments on the LLaMA series, Qwen, and OPT models demonstrate that PermLLM achieves superior performance in optimizing N:M sparse models. The code is available at https://github.com/lanchengzou/PermLLM.

Zehua Pei, Ying Zhang, Hui-Ling Zhen et al.

Mixture-of-experts (MoE) architectures enable scaling large language models (LLMs) to vast parameter counts without a proportional rise in computational costs. However, the significant memory demands of large MoE models hinder their deployment across various computational environments, from cloud servers to consumer devices. This study first demonstrates pronounced task-specific specialization in expert activation patterns within MoE layers. Building on this, we introduce PreMoe, a novel framework that enables efficient deployment of massive MoE models in memory-constrained environments. PreMoe features two main components: probabilistic expert pruning (PEP) and task-adaptive expert retrieval (TAER). PEP employs a new metric, the task-conditioned expected selection score (TCESS), derived from router logits to quantify expert importance for specific tasks, thereby identifying a minimal set of critical experts. TAER leverages these task-specific expert importance profiles for efficient inference. It pre-computes and stores compact expert patterns for diverse tasks. When a user query is received, TAER rapidly identifies the most relevant stored task pattern and reconstructs the model by loading only the small subset of experts crucial for that task. This approach dramatically reduces the memory footprint across all deployment scenarios. DeepSeek-R1 671B maintains 97.2\% accuracy on MATH500 when pruned to 8/128 configuration (50\% expert reduction), and still achieves 72.0\% with aggressive 8/32 pruning (87.5\% expert reduction). Pangu-Ultra-MoE 718B achieves 97.15\% on MATH500 and 81.3\% on AIME24 with 8/128 pruning, while even more aggressive pruning to 4/64 (390GB memory) preserves 96.95\% accuracy on MATH500. We make our code publicly available at https://github.com/JarvisPei/PreMoe.

Juncheng Li, Zehua Pei, Wenjie Li et al.

Single-image super-resolution (SISR) is an important task in image processing, which aims to enhance the resolution of imaging systems. Recently, SISR has made a huge leap and has achieved promising results with the help of deep learning (DL). In this survey, we give an overview of DL-based SISR methods and group them according to their design targets. Specifically, we first introduce the problem definition, research background, and the significance of SISR. Secondly, we introduce some related works, including benchmark datasets, upsampling methods, optimization objectives, and image quality assessment methods. Thirdly, we provide a detailed investigation of SISR and give some domain-specific applications of it. Fourthly, we present the reconstruction results of some classic SISR methods to intuitively know their performance. Finally, we discuss some issues that still exist in SISR and summarize some new trends and future directions. This is an exhaustive survey of SISR, which can help researchers better understand SISR and inspire more exciting research in this field. An investigation project for SISR is provided at https://github.com/CV-JunchengLi/SISR-Survey.

Zehua Pei, Hui-Ling Zhen, Xianzhi Yu et al.

Generative Pre-trained Transformers (GPTs) have demonstrated remarkable performance across diverse domains, largely due to the extensive scaling of model parameters. Recent works have observed redundancy within transformer blocks and developed compression methods by structured pruning of less important blocks. However, such direct removal often leads to irreversible performance degradation. In this paper, we propose FuseGPT, a novel methodology designed to recycle pruned transformer blocks, thereby recovering the model's performance. Firstly, we introduce a new importance detection metric, Macro Influence (MI), which evaluates the long-term impact of each transformer block by quantifying the information loss incurred upon its removal. Next, we propose group-level layer fusion, which leverages the parameters from layers of less important blocks and integrates them into the corresponding layers of neighboring blocks. This fusion process is not a one-time operation but is refined through iterative parameter updates by lightweight group-level fine-tuning. Specifically, the injected parameters are frozen but are weighted with learnable rank decomposition matrices to reduce the computational overhead during fine-tuning. Our approach not only works well for large language models but also for large multimodal models. Experimental results indicate that, even with modest amounts of data, FuseGPT surpasses previous methods in both perplexity and zero-shot task performance.

Yu Zhang, Hui-Ling Zhen, Zehua Pei et al.

Considering the challenges faced by large language models (LLMs) in logical reasoning and planning, prior efforts have sought to augment LLMs with access to external solvers. While progress has been made on simple reasoning problems, solving classical constraint satisfaction problems, such as the Boolean Satisfiability Problem (SAT) and Graph Coloring Problem (GCP), remains difficult for off-the-shelf solvers due to their intricate expressions and exponential search spaces. In this paper, we propose a novel differential logic layer-aided language modeling (DiLA) approach, where logical constraints are integrated into the forward and backward passes of a network layer, to provide another option for LLM tool learning. In DiLA, LLM aims to transform the language description to logic constraints and identify initial solutions of the highest quality, while the differential logic layer focuses on iteratively refining the LLM-prompted solution. Leveraging the logic layer as a bridge, DiLA enhances the logical reasoning ability of LLMs on a range of reasoning problems encoded by Boolean variables, guaranteeing the efficiency and correctness of the solution process. We evaluate the performance of DiLA on two classic reasoning problems and empirically demonstrate its consistent outperformance against existing prompt-based and solver-aided approaches.