Dianbo Sui

Shukang Yin, Chaoyou Fu, Sirui Zhao et al.

Hallucination is a big shadow hanging over the rapidly evolving Multimodal Large Language Models (MLLMs), referring to the phenomenon that the generated text is inconsistent with the image content. In order to mitigate hallucinations, existing studies mainly resort to an instruction-tuning manner that requires retraining the models with specific data. In this paper, we pave a different way, introducing a training-free method named Woodpecker. Like a woodpecker heals trees, it picks out and corrects hallucinations from the generated text. Concretely, Woodpecker consists of five stages: key concept extraction, question formulation, visual knowledge validation, visual claim generation, and hallucination correction. Implemented in a post-remedy manner, Woodpecker can easily serve different MLLMs, while being interpretable by accessing intermediate outputs of the five stages. We evaluate Woodpecker both quantitatively and qualitatively and show the huge potential of this new paradigm. On the POPE benchmark, our method obtains a 30.66%/24.33% improvement in accuracy over the baseline MiniGPT-4/mPLUG-Owl. The source code is released at https://github.com/BradyFU/Woodpecker.

Bozhong Tian, Xiaozhuan Liang, Siyuan Cheng et al.

Large Language Models (LLMs) trained on extensive corpora inevitably retain sensitive data, such as personal privacy information and copyrighted material. Recent advancements in knowledge unlearning involve updating LLM parameters to erase specific knowledge. However, current unlearning paradigms are mired in vague forgetting boundaries, often erasing knowledge indiscriminately. In this work, we introduce KnowUnDo, a benchmark containing copyrighted content and user privacy domains to evaluate if the unlearning process inadvertently erases essential knowledge. Our findings indicate that existing unlearning methods often suffer from excessive unlearning. To address this, we propose a simple yet effective method, MemFlex, which utilizes gradient information to precisely target and unlearn sensitive parameters. Experimental results show that MemFlex is superior to existing methods in both precise knowledge unlearning and general knowledge retaining of LLMs. Code and dataset are released at https://github.com/zjunlp/KnowUnDo.

Zhanyue Qin, Jia Feng, Yibo Lyu et al.

Code reasoning refers to the task of predicting the output of a program given its source code and specific inputs. It can measure the reasoning capability of large language models (LLMs) and also benefit downstream tasks such as code generation and mathematical reasoning. Existing work has verified the effectiveness of reinforcement learning on the task. However, these methods design rewards solely based on final outputs or coarse-grained signals, and neglect the inherent consistency of the stepwise reasoning process in the task. Therefore, these methods often result in sparse reward or reward hacking, which limits the full play of enhanced learning capabilities. To alleviate these issues, we propose CodeThinker, a consistency-driven reinforcement learning framework for code reasoning. Specifically, CodeThinker has three key components: (1) a stepwise reasoning-aware model training module, which utilizes a consistency tracing paradigm as a template to synthesize training data that captures the stepwise reasoning process; (2) a dynamic beam sampling strategy, which aims to improve the quality of sampled outputs under a fixed sampling budget; and (3) a consistency reward mechanism that can effectively alleviate reward hacking. Experiments on three popular benchmarks show that CodeThinker achieves state-of-the-art performance across multiple LLMs. For instance, it outperforms the strongest baseline by 4.3% in accuracy when deployed on Qwen2.5-Coder-7B-Instruct. We also validate the effectiveness of CodeThinker on downstream tasks. Results show that, without additional training, CodeThinker obtains average accuracy gains of 5.33 and 3.11 percentage points on mathematical reasoning and code reasoning tasks covering 17 programming languages, respectively.

Xinye Li, Zunwen Zheng, Qian Zhang et al.

Knowledge Editing (KE) has gained increasing attention, yet current KE tasks remain relatively simple. Under current evaluation frameworks, many editing methods achieve exceptionally high scores, sometimes nearing perfection. However, few studies integrate KE into real-world application scenarios (e.g., recent interest in LLM-as-agent). To support our analysis, we introduce a novel script-based benchmark -- ScEdit (Script-based Knowledge Editing Benchmark) -- which encompasses both counterfactual and temporal edits. We integrate token-level and text-level evaluation methods, comprehensively analyzing existing KE techniques. The benchmark extends traditional fact-based ("What"-type question) evaluation to action-based ("How"-type question) evaluation. We observe that all KE methods exhibit a drop in performance on established metrics and face challenges on text-level metrics, indicating a challenging task. Our benchmark is available at https://github.com/asdfo123/ScEdit.

Bolin Zhang, Zhiwei Yi, Jiahao Wang et al.

The unique diagnosis and treatment techniques and remarkable clinical efficacy of traditional Chinese medicine (TCM) make it play an important role in the field of elderly care and healthcare, especially in the rehabilitation of some common chronic diseases of the elderly. Therefore, building a TCM chatbot for healthcare application will help users obtain consultation services in a direct and natural way. However, concepts such as acupuncture points (acupoints) and meridians involved in TCM always appear in the consultation, which cannot be displayed intuitively. To this end, we develop a \textbf{h}ealthcare chat\textbf{bot} (HBot) based on a human body model in 3D and knowledge graph, which provides conversational services such as knowledge Q\&A, prescription recommendation, moxibustion therapy recommendation, and acupoint search. When specific acupoints are involved in the conversations between user and HBot, the 3D body will jump to the corresponding acupoints and highlight them. Moreover, Hbot can also be used in training scenarios to accelerate the teaching process of TCM by intuitively displaying acupuncture points and knowledge cards. The demonstration video is available at https://www.youtube.com/watch?v=UhQhutSKkTU . Our code and dataset are publicly available at Gitee: https://gitee.com/plabrolin/interactive-3d-acup.git

Deyuan Liu, Zecheng Wang, Zhanyue Qin et al.

Large Language Models (LLMs) exhibit impressive capabilities yet suffer from sensitivity to slight input context variations, hampering reliability. Conventional metrics like accuracy and perplexity fail to assess local prediction robustness, as normalized output probabilities can obscure the underlying resilience of an LLM's internal state to perturbations. We introduce the Token Constraint Bound ($δ_{\mathrm{TCB}}$), a novel metric that quantifies the maximum internal state perturbation an LLM can withstand before its dominant next-token prediction significantly changes. Intrinsically linked to output embedding space geometry, $δ_{\mathrm{TCB}}$ provides insights into the stability of the model's internal predictive commitment. Our experiments show $δ_{\mathrm{TCB}}$ correlates with effective prompt engineering and uncovers critical prediction instabilities missed by perplexity during in-context learning and text generation. $δ_{\mathrm{TCB}}$ offers a principled, complementary approach to analyze and potentially improve the contextual stability of LLM predictions.

Xinye Li, Mingqi Wan, Dianbo Sui

We present Team asdfo123's submission to the LLMSR@XLLM25 shared task, which evaluates large language models on producing fine-grained, controllable, and interpretable reasoning processes. Systems must extract all problem conditions, decompose a chain of thought into statement-evidence pairs, and verify the logical validity of each pair. Leveraging only the off-the-shelf Meta-Llama-3-8B-Instruct, we craft a concise few-shot, multi-turn prompt that first enumerates all conditions and then guides the model to label, cite, and adjudicate every reasoning step. A lightweight post-processor based on regular expressions normalises spans and enforces the official JSON schema. Without fine-tuning, external retrieval, or ensembling, our method ranks 5th overall, achieving macro F1 scores on par with substantially more complex and resource-consuming pipelines. We conclude by analysing the strengths and limitations of our approach and outlining directions for future research in structural reasoning with LLMs. Our code is available at https://github.com/asdfo123/LLMSR-asdfo123.

Dianbo Sui, Yubo Chen, Kang Liu et al.

The joint entity and relation extraction task aims to extract all relational triples from a sentence. In essence, the relational triples contained in a sentence are unordered. However, previous seq2seq based models require to convert the set of triples into a sequence in the training phase. To break this bottleneck, we treat joint entity and relation extraction as a direct set prediction problem, so that the extraction model can get rid of the burden of predicting the order of multiple triples. To solve this set prediction problem, we propose networks featured by transformers with non-autoregressive parallel decoding. Unlike autoregressive approaches that generate triples one by one in a certain order, the proposed networks directly output the final set of triples in one shot. Furthermore, we also design a set-based loss that forces unique predictions via bipartite matching. Compared with cross-entropy loss that highly penalizes small shifts in triple order, the proposed bipartite matching loss is invariant to any permutation of predictions; thus, it can provide the proposed networks with a more accurate training signal by ignoring triple order and focusing on relation types and entities. Experiments on two benchmark datasets show that our proposed model significantly outperforms current state-of-the-art methods. Training code and trained models will be available at http://github.com/DianboWork/SPN4RE.

Yongxin Guo, Jingyu Liu, Mingda Li et al.

Video Temporal Grounding (VTG) strives to accurately pinpoint event timestamps in a specific video using linguistic queries, significantly impacting downstream tasks like video browsing and editing. Unlike traditional task-specific models, Video Large Language Models (video LLMs) can handle multiple tasks concurrently in a zero-shot manner. Consequently, exploring the application of video LLMs for VTG tasks has become a burgeoning research area. However, despite considerable advancements in video content understanding, video LLMs often struggle to accurately pinpoint timestamps within videos, limiting their effectiveness in VTG tasks. To address this, we introduce VTG-LLM, a model designed to enhance video LLMs' timestamp localization abilities. Our approach includes: (1) effectively integrating timestamp knowledge into visual tokens; (2) incorporating absolute-time tokens to manage timestamp knowledge without concept shifts; and (3) introducing a lightweight, high-performance, slot-based token compression technique designed to accommodate the demands of a large number of frames to be sampled for VTG tasks. Additionally, we present VTG-IT-120K, a collection of publicly available VTG datasets that we have re-annotated to improve upon low-quality annotations. Our comprehensive experiments demonstrate the superior performance of VTG-LLM in comparison to other video LLM methods across a variety of VTG tasks.

Deyuan Liu, Zecheng Wang, Bingning Wang et al.

The rapid proliferation of large language models (LLMs) such as GPT-4 and Gemini underscores the intense demand for resources during their training processes, posing significant challenges due to substantial computational and environmental costs. To alleviate this issue, we propose checkpoint merging in pretraining LLM. This method utilizes LLM checkpoints with shared training trajectories, and is rooted in an extensive search space exploration for the best merging weight via Bayesian optimization. Through various experiments, we demonstrate that: (1) Our proposed methodology exhibits the capacity to augment pretraining, presenting an opportunity akin to obtaining substantial benefits at minimal cost; (2) Our proposed methodology, despite requiring a given held-out dataset, still demonstrates robust generalization capabilities across diverse domains, a pivotal aspect in pretraining.

Haochuan Wang, Xiachong Feng, Lei Li et al. · pku

The rapid advancement of large language models has accelerated their application in reasoning, with strategic reasoning drawing increasing attention. To evaluate the strategic reasoning capabilities of LLMs, game theory, with its concise structure, has become the preferred approach for many researchers. However, current research typically focuses on a limited selection of games, resulting in low coverage of game types. Additionally, classic game scenarios carry risks of data leakage, and the benchmarks used often lack extensibility, rendering them inadequate for evaluating state-of-the-art models. To address these challenges, we propose TMGBench, characterized by comprehensive game type coverage, diverse scenarios and flexible game organization. Specifically, we incorporate all 144 game types summarized by the Robinson-Goforth topology of 2x2 games, constructed as classic games in our benchmark; we also synthetize diverse, higher-quality game scenarios for each classic game, which we refer to as story-based games. Lastly, to provide a sustainable evaluation framework adaptable to increasingly powerful LLMs, we treat the aforementioned games as atomic units and organize them into more complex forms through sequential, parallel, and nested structures. We conducted a comprehensive evaluation of mainstream LLMs, covering tests on rational reasoning, reasoning robustness, Theory-of-Mind capabilities, and reasoning in complex game forms. The results revealed LLMs still have flaws in the accuracy and consistency of strategic reasoning processes, and their levels of mastery over Theory-of-Mind also vary. Additionally, SOTA models like o3-mini, Qwen3 and deepseek-reasoner, were also evaluated across the sequential, parallel, and nested game structures while the results highlighted the challenges posed by TMGBench.

Zecheng Wang, Deyuan Liu, Chunshan Li et al.

Standard negative log-likelihood (NLL) for Supervised Fine-Tuning (SFT) applies uniform token-level weighting. This rigidity creates a two-fold failure mode: (i) overemphasizing low-probability targets can amplify gradients on noisy supervision and disrupt robust priors, and (ii) uniform weighting provides weak sharpening when the model is already confident. Existing methods fail to resolve the resulting plasticity--stability dilemma, often suppressing necessary learning signals alongside harmful ones. To address this issue, we unify token-level SFT objectives within a generalized deformed-log family and expose a universal gate $\times$ error gradient structure, where the gate controls how much the model trusts its current prediction. By employing the Cayley transform, we map the model's continuously evolving uncertainty onto a continuous focus trajectory, which enables seamless interpolation between scenarios involving uncertain novel concepts and those involving well-established knowledge. We then introduce Dynamic Entropy Fine-Tuning (DEFT), a parameter-free objective that modulates the trust gate using distribution concentration (Rényi-2 entropy) as a practical proxy for the model's predictive state. Extensive experiments and analyses demonstrate that DEFT achieves a better balance between exploration and exploitation, leading to improved overall performance.

Rihui Xin, Han Liu, Zecheng Wang et al.

Large Language Models have achieved remarkable success in natural language processing tasks, with Reinforcement Learning playing a key role in adapting them to specific applications. However, obtaining ground truth answers for training LLMs in mathematical problem-solving is often challenging, costly, and sometimes unfeasible. This research delves into the utilization of format and length as surrogate signals to train LLMs for mathematical problem-solving, bypassing the need for traditional ground truth answers. Our study shows that a reward function centered on format correctness alone can yield performance improvements comparable to the standard GRPO algorithm in early phases. Recognizing the limitations of format-only rewards in the later phases, we incorporate length-based rewards. The resulting GRPO approach, leveraging format-length surrogate signals, not only matches but surpasses the performance of the standard GRPO algorithm relying on ground truth answers in certain scenarios, achieving 40.0% accuracy on AIME2024 with a 7B base model. Through systematic exploration and experimentation, this research not only offers a practical solution for training LLMs to solve mathematical problems and reducing the dependence on extensive ground truth data collection, but also reveals the essence of why our label-free approach succeeds: the powerful base model is like an excellent student who has already mastered mathematical and logical reasoning skills, but performs poorly on the test paper, it simply needs to develop good answering habits to achieve outstanding results in exams, to unlock the capabilities it already possesses.

Zhanyue Qin, Yue Ding, Deyuan Liu et al.

Nowadays, Large Language Models (LLMs) have attracted widespread attention due to their powerful performance. However, due to the unavoidable exposure to socially biased data during training, LLMs tend to exhibit social biases, particularly gender bias. To better explore and quantifying the degree of gender bias in LLMs, we propose a pair of datasets named GenBiasEval and GenHintEval, respectively. The GenBiasEval is responsible for evaluating the degree of gender bias in LLMs, accompanied by an evaluation metric named AFGB-Score (Absolutely Fair Gender Bias Score). Meanwhile, the GenHintEval is used to assess whether LLMs can provide responses consistent with prompts that contain gender hints, along with the accompanying evaluation metric UB-Score (UnBias Score). Besides, in order to mitigate gender bias in LLMs more effectively, we present the LFTF (Locating First and Then Fine-Tuning) algorithm.The algorithm first ranks specific LLM blocks by their relevance to gender bias in descending order using a metric called BMI (Block Mitigating Importance Score). Based on this ranking, the block most strongly associated with gender bias is then fine-tuned using a carefully designed loss function. Numerous experiments have shown that our proposed LFTF algorithm can significantly mitigate gender bias in LLMs while maintaining their general capabilities.

Zhanyue Qin, Haochuan Wang, Deyuan Liu et al.

Sequential decision-making refers to algorithms that take into account the dynamics of the environment, where early decisions affect subsequent decisions. With large language models (LLMs) demonstrating powerful capabilities between tasks, we can't help but ask: Can Current LLMs Effectively Make Sequential Decisions? In order to answer this question, we propose the UNO Arena based on the card game UNO to evaluate the sequential decision-making capability of LLMs and explain in detail why we choose UNO. In UNO Arena, We evaluate the sequential decision-making capability of LLMs dynamically with novel metrics based Monte Carlo methods. We set up random players, DQN-based reinforcement learning players, and LLM players (e.g. GPT-4, Gemini-pro) for comparison testing. Furthermore, in order to improve the sequential decision-making capability of LLMs, we propose the TUTRI player, which can involves having LLMs reflect their own actions wtih the summary of game history and the game strategy. Numerous experiments demonstrate that the TUTRI player achieves a notable breakthrough in the performance of sequential decision-making compared to the vanilla LLM player.

Deyuan Liu, Zhanyue Qin, Hairu Wang et al.

While large language models (LLMs) excel in many domains, their complexity and scale challenge deployment in resource-limited environments. Current compression techniques, such as parameter pruning, often fail to effectively utilize the knowledge from pruned parameters. To address these challenges, we propose Manifold-Based Knowledge Alignment and Layer Merging Compression (MKA), a novel approach that uses manifold learning and the Normalized Pairwise Information Bottleneck (NPIB) measure to merge similar layers, reducing model size while preserving essential performance. We evaluate MKA on multiple benchmark datasets and various LLMs. Our findings show that MKA not only preserves model performance but also achieves substantial compression ratios, outperforming traditional pruning methods. Moreover, when coupled with quantization, MKA delivers even greater compression. Specifically, on the MMLU dataset using the Llama3-8B model, MKA achieves a compression ratio of 43.75% with a minimal performance decrease of only 2.82\%. The proposed MKA method offers a resource-efficient and performance-preserving model compression technique for LLMs.

Dianbo Sui, Yubo Chen, Kang Liu et al.

This paper investigates distantly supervised relation extraction in federated settings. Previous studies focus on distant supervision under the assumption of centralized training, which requires collecting texts from different platforms and storing them on one machine. However, centralized training is challenged by two issues, namely, data barriers and privacy protection, which make it almost impossible or cost-prohibitive to centralize data from multiple platforms. Therefore, it is worthy to investigate distant supervision in the federated learning paradigm, which decouples the model training from the need for direct access to the raw data. Overcoming label noise of distant supervision, however, becomes more difficult in federated settings, since the sentences containing the same entity pair may scatter around different platforms. In this paper, we propose a federated denoising framework to suppress label noise in federated settings. The core of this framework is a multiple instance learning based denoising method that is able to select reliable instances via cross-platform collaboration. Various experimental results on New York Times dataset and miRNA gene regulation relation dataset demonstrate the effectiveness of the proposed method.

Dianbo Sui, Yubo Chen, Binjie Mao et al.

The training of deep-learning-based text classification models relies heavily on a huge amount of annotation data, which is difficult to obtain. When the labeled data is scarce, models tend to struggle to achieve satisfactory performance. However, human beings can distinguish new categories very efficiently with few examples. This is mainly due to the fact that human beings can leverage knowledge obtained from relevant tasks. Inspired by human intelligence, we propose to introduce external knowledge into few-shot learning to imitate human knowledge. A novel parameter generator network is investigated to this end, which is able to use the external knowledge to generate relation network parameters. Metrics can be transferred among tasks when equipped with these generated parameters, so that similar tasks use similar metrics while different tasks use different metrics. Through experiments, we demonstrate that our method outperforms the state-of-the-art few-shot text classification models.