Zhengze Zhang

Zuyao Xu, Yuqi Qiu, Lu Sun et al.

Citations provide the basis for trusting scientific claims; when they are invalid or fabricated, this trust collapses. With the advent of Large Language Models (LLMs), this risk has intensified: LLMs are increasingly used for academic writing, yet their tendency to fabricate citations (``ghost citations'') poses a systemic threat to citation validity. To quantify this threat and inform mitigation, we develop CiteVerifier, an open-source framework for large-scale citation verification, and conduct the first comprehensive study of citation validity in the LLM era through three experiments built on it. We benchmark 13 state-of-the-art LLMs on citation generation across 40 research domains, finding that all models hallucinate citations at rates from 14.23\% to 94.93\%, with significant variation across research domains. Moreover, we analyze 2.2 million citations from 56,381 papers published at top-tier AI/ML and Security venues (2020--2025), confirming that 1.07\% of papers contain invalid or fabricated citations (604 papers), with an 80.9\% increase in 2025 alone. Furthermore, we survey 97 researchers and analyze 94 valid responses after removing 3 conflicting samples, revealing a critical ``verification gap'': 41.5\% of researchers copy-paste BibTeX without checking and 44.4\% choose no-action responses when encountering suspicious references; meanwhile, 76.7\% of reviewers do not thoroughly check references and 80.0\% never suspect fake citations. Our findings reveal an accelerating crisis where unreliable AI tools, combined with inadequate human verification by researchers and insufficient peer review scrutiny, enable fabricated citations to contaminate the scientific record. We propose interventions for researchers, venues, and tool developers to protect citation integrity.

Shiqi Wang, Zhengze Zhang, Rui Zhao et al.

With the rapid advances in Large Language Models (LLMs), aligning LLMs with human preferences become increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the "ordinal relationship" between responses, overlooking the crucial aspect of how much one is preferred over the others. To address this issue, we propose a simple yet effective solution called Reward Difference Optimization, shorted as RDO. Specifically, we introduce reward difference coefficients to reweigh sample pairs in offline RLHF. We then develop a difference model which captures rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR datasets substantiate the effectiveness of our method in both automatic metrics and human evaluation, thereby highlighting its potential for aligning LLMs with human intent and values

Yiqun Shen, Song Yuan, Zhengze Zhang et al.

KV Cache is commonly used to accelerate LLM inference with long contexts, yet its high memory demand drives the need for cache compression. Existing compression methods, however, are largely heuristic and lack dynamic budget allocation. To address this limitation, we introduce a unified framework for cache compression by minimizing information loss in Transformer residual streams. Building on it, we analyze the layer attention output loss and derive a new metric to compare cache entries across heads, enabling layer-wise compression with dynamic head budgets. Additionally, by contrasting cross-layer information, we also achieve dynamic layer budgets. LAVa is the first unified strategy for cache eviction and dynamic budget allocation that, unlike prior methods, does not rely on training or the combination of multiple strategies. Experiments with benchmarks (LongBench, Needle-In-A-Haystack, Ruler, and InfiniteBench) demonstrate its superiority. Moreover, our experiments reveal a new insight: dynamic layer budgets are crucial for generation tasks (e.g., code completion), while dynamic head budgets play a key role in extraction tasks (e.g., extractive QA). As a fully dynamic compression method, LAVa consistently maintains top performance across task types. Our code is available at https://github.com/MGDDestiny/Lava.

Zhengze Zhang, Shiqi Wang, Yiqun Shen et al.

Large language models (LLMs) have demonstrated exceptional performance across various applications, but their conversational abilities decline sharply as model size decreases, presenting a barrier to their deployment in resource-constrained environments. Knowledge distillation with Direct Preference Optimization (dDPO) has emerged as a promising approach to enhancing the conversational abilities of smaller models using a larger teacher model. However, current methods primarily focus on 'black-box' KD, which only uses the teacher's responses, overlooking the output distribution offered by the teacher. This paper addresses this gap by introducing daDPO (Distribution-Aware DPO), a unified method for preference optimization and distribution-based distillation. We provide rigorous theoretical analysis and empirical validation, showing that daDPO outperforms existing methods in restoring performance for pruned models and enhancing smaller LLM models. Notably, in in-domain evaluation, our method enables a 20% pruned Vicuna1.5-7B to achieve near-teacher performance (-7.3% preference rate compared to that of dDPO's -31%), and allows Qwen2.5-1.5B to occasionally outperform its 7B teacher model (14.0% win rate).