arXiv subjects

Shuaishuai Zu

Publications and source records attributed to Shuaishuai Zu.

8 recordsLinked to original sources

arXiv2026

O$^2$PD: Overlap-Gated On-Policy Distillation on Reliable Prefix

On-policy distillation (OPD) supplies dense token-level supervision by aligning a student to the teacher's next-token distribution on its own generated rollouts. Existing works mostly implement top-$k$ OPD, applying reverse Kullback-Leibler ($\RKL$) divergence between the teacher's and the student's renormalized distributions over teacher top-$k$ support. However, the teacher scores student-generated rollouts that are inherently off-policy for itself, leaving the reliability of its next-token guidance unclear. Restricting the divergence to the teacher top-$k$ support also discards the remaining mass outside that support. These gaps leave two questions implicit: 1) which tokens to align, and 2) how to align those tokens. To answer these questions, we propose Overlap-gated On-Policy Distillation (O$^{2}$PD). As for 1), we quantitatively analyze the reliability of teacher supervision and find that alignment on the student's trajectory prefix is more reliable than alignment on the full context. Therefore, our O$^{2}$PD is applied only to the 20\% prefix tokens of each student rollout. As for 2), we theoretically characterize two problems of top-$k$ OPD: the \emph{Teacher-support Uncovering} and the \emph{Dominant-Mode Mismatch}. Based on these analyses, we propose two support strategies: the \emph{Reset Support} strategy attaches a reset atom to every token, pooling mass outside the teacher top-$k$, and the \emph{Support Convergence} strategy converges the support to the already aligned intersection. Our O$^{2}$PD employs the designed overlap-gated $α$-divergence to interpolate between mode-covering and mode-seeking for per-token alignment, where the overlap is used as a gate to switch the support and adaptively modulate $α$.

cs.CL↗

arXiv2026

TraceSpec: Understanding and Improving Speculative Decoding for LLM Agents

LLM agents execute tasks through repeated model calls over evolving histories of reasoning, actions, and tool feedback. Aggregate speculative-decoding metrics merge these distinct decisions, providing limited guidance on where draft supervision is needed. We present TraceSpec, a framework connecting trajectory evaluation with adaptive draft training. It combines dynamic execution with fixed-prefix replay and introduces TraceSpec-Traj, a corpus of 10,000 training states and 2,500 test states spanning browsing and software engineering. Analysis of five speculative baselines reveals that aggregate acceptance can rise with context length while declining within matched late browsing actions. Workload and phase further distinguish these trends, motivating supervision guided by trajectory conditions. TraceSpec trains on intermediate assistant turns and allocates a fixed prediction budget using state-conditioned estimates of prefix-aware acceptance gaps. Across three target sizes and two temperatures, it improves replay throughput by 12.5% to 22.6% over its DSpark base. Qwen3-4B ablations show that adaptive allocation increases accepted length by 3.3% to 10.0% over fixed full-turn supervision, with additional comparisons substantiating the roles of state conditioning and consecutive-token agreement. The resulting evaluation resources, trajectory findings, and training strategy connect the analysis of agent execution to practical draft improvement.

cs.CL↗

arXiv2026

Writer-R1: Dynamic Rubric Mining for Reinforcement Learning of Open-Ended Writing

Training models for open-ended writing requires diverse prompts and sufficiently precise specifications of response quality. Constructing such specifications at scale entails recovering task requirements, consolidating overlapping criteria, and defining observable differences between performance levels. We introduce Writer-R1, a writing training pipeline built around Inductive Rubric Generation (IRG), which integrates these operations into an automated rubric construction procedure. The pipeline expands an open writing corpus and uses the resulting prompt-specific rubrics throughout supervised training and reinforcement learning (RL). Each rubric is concatenated with the user request and also defines the reward criteria, making the requirements used for assessment available to the writer before generation. A compact model, RubricGen-4B, learns to supply rubrics for new requests. Writer-R1 improves WritingBench scores by 0.182 and 0.158 over the 4B and 8B backbones, respectively, with gains also on HelloBench. Training ablations favor IRG over direct rubric generation, and further analyses support rubric transfer and the robustness of writer improvements across evaluators. These results support using inductively mined rubrics as both generation guidance and reward specifications for open-ended writing.

cs.CL↗

arXiv2026

TreeSpark: Learning to Draft the Trees You Verify for Speculative Decoding

Parallel drafters such as DFlash and DSpark reduce proposal latency, but early mismatches limit how much of each block can be accepted. Retaining alternatives introduces parent states and candidate choices that reference-conditioned training does not directly supervise. We introduce TreeSpark, a unified training and inference framework extending the semi-autoregressive DSpark architecture. It constructs parent-conditioned candidate trees from one backbone pass and replays their expansion, pruning, and budgeted packing during training. Survival negative log-likelihood weights retained edges by reference-derived path mass, while a capped coverage hinge promotes target-important alternatives near local selection boundaries. Both objectives reuse reference-chain target distributions without additional target forward passes. We evaluate greedy decoding with Qwen3-4B and Qwen3-8B on nine general benchmarks and six writing domains. Across targets and suites, TreeSpark improves mean acceptance length over DSpark by 23.4% to 31.6% and mean single-request throughput by 15.1% to 25.0%. Batched Qwen3-4B throughput improves by 12.6% to 17.2%. At fixed attention capacity and inference policy, the proposed supervision further improves mean acceptance length by approximately 6.4% and mean system throughput by 4.2% to 5.0% on Qwen3-4B.

cs.CL↗

arXiv2026

QChunker: Learning Question-Aware Text Chunking for Domain RAG via Multi-Agent Debate

The effectiveness upper bound of retrieval-augmented generation (RAG) is fundamentally constrained by the semantic integrity and information granularity of text chunks in its knowledge base. To address these challenges, this paper proposes QChunker, which restructures the RAG paradigm from retrieval-augmentation to understanding-retrieval-augmentation. Firstly, QChunker models the text chunking as a composite task of text segmentation and knowledge completion to ensure the logical coherence and integrity of text chunks. Drawing inspiration from Hal Gregersen's "Questions Are the Answer" theory, we design a multi-agent debate framework comprising four specialized components: a question outline generator, text segmenter, integrity reviewer, and knowledge completer. This framework operates on the principle that questions serve as catalysts for profound insights. Through this pipeline, we successfully construct a high-quality dataset of 45K entries and transfer this capability to small language models. Additionally, to handle long evaluation chains and low efficiency in existing chunking evaluation methods, which overly rely on downstream QA tasks, we introduce a novel direct evaluation metric, ChunkScore. Both theoretical and experimental validations demonstrate that ChunkScore can directly and efficiently discriminate the quality of text chunks. Furthermore, during the text segmentation phase, we utilize document outlines for multi-path sampling to generate multiple candidate chunks and select the optimal solution employing ChunkScore. Extensive experimental results across four heterogeneous domains exhibit that QChunker effectively resolves aforementioned issues by providing RAG with more logically coherent and information-rich text chunks.

cs.CL↗

arXiv2025

Invoke Interfaces Only When Needed: Adaptive Invocation for Large Language Models in Question Answering

The collaborative paradigm of large and small language models (LMs) effectively balances performance and cost, yet its pivotal challenge lies in precisely pinpointing the moment of invocation when hallucinations arise in small LMs. Previous optimization efforts primarily focused on post-processing techniques, which were separate from the reasoning process of LMs, resulting in high computational costs and limited effectiveness. In this paper, we propose a practical invocation evaluation metric called AttenHScore, which calculates the accumulation and propagation of hallucinations during the generation process of small LMs, continuously amplifying potential reasoning errors. By dynamically adjusting the detection threshold, we achieve more accurate real-time invocation of large LMs. Additionally, considering the limited reasoning capacity of small LMs, we leverage uncertainty-aware knowledge reorganization to assist them better capture critical information from different text chunks. Extensive experiments reveal that our AttenHScore outperforms most baselines in enhancing real-time hallucination detection capabilities across multiple QA datasets, especially when addressing complex queries. Moreover, our strategies eliminate the need for additional model training and display flexibility in adapting to various transformer-based LMs.

cs.CL↗

arXiv2025

RouterKT: Mixture-of-Experts for Knowledge Tracing

Knowledge Tracing (KT) is a fundamental task in Intelligent Tutoring Systems (ITS), which aims to model the dynamic knowledge states of students based on their interaction histories. However, existing KT models often rely on a global forgetting decay mechanism for capturing learning patterns, assuming that students' performance is predominantly influenced by their most recent interactions. Such approaches fail to account for the diverse and complex learning patterns arising from individual differences and varying learning stages. To address this limitation, we propose RouterKT, a novel Mixture-of-Experts (MoE) architecture designed to capture heterogeneous learning patterns by enabling experts to specialize in different patterns without any handcrafted learning pattern bias such as forgetting decay. Specifically, RouterKT introduces a \textbf{person-wise routing mechanism} to effectively model individual-specific learning behaviors and employs \textbf{multi-heads as experts} to enhance the modeling of complex and diverse patterns. Comprehensive experiments on ten benchmark datasets demonstrate that RouterKT exhibits significant flexibility and improves the performance of various KT backbone models, with a maximum average AUC improvement of 3.29\% across different backbones and datasets, outperforming other state-of-the-art models. Moreover, RouterKT demonstrates consistently superior inference efficiency compared to existing approaches based on handcrafted learning pattern bias, highlighting its usability for real-world educational applications. The source code is available at https://github.com/ringotc/RouterKT.git.

cs.CY↗

arXiv2024

Fair Knowledge Tracing in Second Language Acquisition

In second-language acquisition, predictive modeling aids educators in implementing diverse teaching strategies, attracting significant research attention. However, while model accuracy is widely explored, model fairness remains under-examined. Model fairness ensures equitable treatment of groups, preventing unintentional biases based on attributes such as gender, ethnicity, or economic background. A fair model should produce impartial outcomes that do not systematically disadvantage any group. This study evaluates the fairness of two predictive models using the Duolingo dataset's en\_es (English learners speaking Spanish), es\_en (Spanish learners speaking English), and fr\_en (French learners speaking English) tracks. We analyze: 1. Algorithmic fairness across platforms (iOS, Android, Web). 2. Algorithmic fairness between developed and developing countries. Key findings include: 1. Deep learning outperforms machine learning in second-language knowledge tracing due to improved accuracy and fairness. 2. Both models favor mobile users over non-mobile users. 3. Machine learning exhibits stronger bias against developing countries compared to deep learning. 4. Deep learning strikes a better balance of fairness and accuracy in the en\_es and es\_en tracks, while machine learning is more suitable for fr\_en. This study highlights the importance of addressing fairness in predictive models to ensure equitable educational strategies across platforms and regions.

cs.HC↗