Wei Li
CS PhD Student at 
Previously Economics BA Student at 
About
Wei Li developed FEA-Bench to evaluate repository-level code generation for implementing software features. Li is a PhD student at Peking University’s Institute of Computational Linguistics, where they are advised by Houfeng Wang. Their research focuses on coding agents and LLMs, specifically optimizing models for long-horizon software engineering tasks and learning from complex problems. Li is currently a research intern at ByteDance Seed working on software engineering data, following a year-long residency at Microsoft Research Asia focused on repository-level coding. They also authored a multimodal sentiment analysis framework using transformer-based feature reconstruction and a detection-correction structure for grammatical error correction. Li previously earned a BE in Automation and a BA in Economics from Tsinghua University.
Experience
CS PhD Student
Aug 2022 – Present
Member of the Institute of Computational Linguistics; advised by Prof. Houfeng Wang.
Aug 2022 – Present
Nov 2025 – Present
Member
Aug 2022 – Present
Institute of Computational Linguistics
Aug 2022 – Present
Sep 2024 – Sep 2025
Economics BA Student
Sep 2019 – Jul 2022
Completed a second bachelor's degree.
Sep 2019 – Jul 2022
Automation BE Student
May 2019 – Jul 2022
May 2019 – Jul 2022
Mechanical Engineering Student
Sep 2017 – May 2019
Studied in the mechanical engineering broad category.
Sep 2017 – May 2019
Research Intern
Jul 2021 – Sep 2021
Recurrent AI
Worked with the MoonShot Team on zero-shot learning and prompt tuning.
Jul 2021 – Sep 2021
Papers14
TIME: A Multi-level Benchmark for Temporal Reasoning of LLMs in Real-World Scenarios
Temporal reasoning is pivotal for Large Language Models (LLMs) to comprehend the real world. However, existing works neglect the real-world challenges for temporal reasoning: (1) intensive temporal information, (2) fast-changing event dynamics, and (3) complex temporal dependencies in social interactions. To bridge this gap, we propose a multi-level benchmark TIME, designed for temporal reasoning in real-world scenarios. TIME consists of 38,522 QA pairs, covering 3 levels with 11 fine-grained sub-tasks. This benchmark encompasses 3 sub-datasets reflecting different real-world challenges: TIME-Wiki, TIME-News, and TIME-Dial. We conduct extensive experiments on reasoning models and non-reasoning models. And we conducted an in-depth analysis of temporal reasoning performance across diverse real-world scenarios and tasks, and summarized the impact of test-time scaling on temporal reasoning capabilities. Additionally, we release TIME-Lite, a human-annotated subset to foster future research and standardized evaluation in temporal reasoning. The code is available at this https URL , the dataset is available at this https URL , and the project page link is this https URL .
HalluDial: A Large-Scale Benchmark for Automatic Dialogue-Level Hallucination Evaluation
Large Language Models (LLMs) have significantly advanced the field of Natural Language Processing (NLP), achieving remarkable performance across diverse tasks and enabling widespread real-world applications. However, LLMs are prone to hallucination, generating content that either conflicts with established knowledge or is unfaithful to the original sources. Existing hallucination benchmarks primarily focus on sentence- or passage-level hallucination detection, neglecting dialogue-level evaluation, hallucination localization, and rationale provision. They also predominantly target factuality hallucinations while underestimating faithfulness hallucinations, often relying on labor-intensive or non-specialized evaluators. To address these limitations, we propose HalluDial, the first comprehensive large-scale benchmark for automatic dialogue-level hallucination evaluation. HalluDial encompasses both spontaneous and induced hallucination scenarios, covering factuality and faithfulness hallucinations. The benchmark includes 4,094 dialogues with a total of 146,856 samples. Leveraging HalluDial, we conduct a comprehensive meta-evaluation of LLMs' hallucination evaluation capabilities in information-seeking dialogues and introduce a specialized judge language model, HalluJudge. The high data quality of HalluDial enables HalluJudge to achieve superior or competitive performance in hallucination evaluation, facilitating the automatic assessment of dialogue-level hallucinations in LLMs and providing valuable insights into this phenomenon. The dataset and the code are available at this https URL.
Odysseus Navigates the Sirens' Song: Dynamic Focus Decoding for Factual and Diverse Open-Ended Text Generation
Large Language Models (LLMs) are increasingly required to generate text that is both factually accurate and diverse across various open-ended applications. However, current stochastic decoding methods struggle to balance such objectives. We introduce Dynamic Focus Decoding (DFD), a novel plug-and-play stochastic approach that resolves this trade-off without requiring additional data, knowledge, or models. DFD adaptively adjusts the decoding focus based on distributional differences across layers, leveraging the modular and hierarchical nature of factual knowledge within LLMs. This dynamic adjustment improves factuality in knowledge-intensive decoding steps and promotes diversity in less knowledge-reliant steps. DFD can be easily integrated with existing decoding methods, enhancing both factuality and diversity with minimal computational overhead. Extensive experiments across seven datasets demonstrate that DFD significantly improves performance, providing a scalable and efficient solution for open-ended text generation.
Two Pathways to Truthfulness: On the Intrinsic Encoding of LLM Hallucinations
Despite their impressive capabilities, large language models (LLMs) frequently generate hallucinations. Previous work shows that their internal states encode rich signals of truthfulness, yet the origins and mechanisms of these signals remain unclear. In this paper, we demonstrate that truthfulness cues arise from two distinct information pathways: (1) a Question-Anchored pathway that depends on question-answer information flow, and (2) an Answer-Anchored pathway that derives self-contained evidence from the generated answer itself. First, we validate and disentangle these pathways through attention knockout and token patching. Afterwards, we uncover notable and intriguing properties of these two mechanisms. Further experiments reveal that (1) the two mechanisms are closely associated with LLM knowledge boundaries; and (2) internal representations are aware of their distinctions. Finally, building on these insightful findings, two applications are proposed to enhance hallucination detection performance. Overall, our work provides new insight into how LLMs internally encode truthfulness, offering directions for more reliable and self-aware generative systems.
FEA-Bench: A Benchmark for Evaluating Repository-Level Code Generation for Feature Implementation
Implementing new features in repository-level codebases is a crucial application of code generation models. However, current benchmarks lack a dedicated evaluation framework for this capability. To fill this gap, we introduce FEA-Bench, a benchmark designed to assess the ability of large language models (LLMs) to perform incremental development within code repositories. We collect pull requests from 83 GitHub repositories and use rule-based and intent-based filtering to construct task instances focused on new feature development. Each task instance containing code changes is paired with relevant unit test files to ensure that the solution can be verified. The feature implementation requires LLMs to simultaneously possess code completion capabilities for new components and code editing abilities for other relevant parts in the code repository, providing a more comprehensive evaluation method of LLMs' automated software engineering capabilities. Experimental results show that LLMs perform significantly worse in the FEA-Bench, highlighting considerable challenges in such repository-level incremental code development.
Mitigating Overthinking through Reasoning Shaping
Large reasoning models (LRMs) boosted by Reinforcement Learning from Verifier Reward (RLVR) have shown great power in problem solving, yet they often cause overthinking: excessive, meandering reasoning that inflates computational cost. Prior designs of penalization in RLVR manage to reduce token consumption while often harming model performance, which arises from the oversimplicity of token-level supervision. In this paper, we argue that the granularity of supervision plays a crucial role in balancing efficiency and accuracy, and propose Group Relative Segment Penalization (GRSP), a step-level method to regularize reasoning. Since preliminary analyses show that reasoning segments are strongly correlated with token consumption and model performance, we design a length-aware weighting mechanism across segment clusters. Extensive experiments demonstrate that GRSP achieves superior token efficiency without heavily compromising accuracy, especially the advantages with harder problems. Moreover, GRSP stabilizes RL training and scales effectively across model sizes.
Detection-Correction Structure via General Language Model for Grammatical Error Correction
Wei LiHoufeng WangGrammatical error correction (GEC) is a task dedicated to rectifying texts with minimal edits, which can be decoupled into two components: detection and correction. However, previous works have predominantly focused on direct correction, with no prior efforts to integrate both into a single model. Moreover, the exploration of the detection-correction paradigm by large language models (LLMs) remains underdeveloped. This paper introduces an integrated detection-correction structure, named DeCoGLM, based on the General Language Model (GLM). The detection phase employs a fault-tolerant detection template, while the correction phase leverages autoregressive mask infilling for localized error correction. Through the strategic organization of input tokens and modification of attention masks, we facilitate multi-task learning within a single model. Our model demonstrates competitive performance against the state-of-the-art models on English and Chinese GEC datasets. Further experiments present the effectiveness of the detection-correction structure in LLMs, suggesting a promising direction for GEC.
Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generation
Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but answer-visible generation can justify a pre-committed answer rather than derive it. This post-hoc rationalization creates a train-inference mismatch because student models are trained on answer-conditioned traces but must reason without answer access at inference time. We quantify this mismatch through lexical, trajectory, and probabilistic anchoring, measuring surface overlap, answer-conditioned generation dynamics, and answer recoverability from the trace, respectively. We find that semantic suppression, a seemingly intuitive mitigation, reduces lexical overlap but increases trajectory anchoring: avoiding the answer requires continually tracking it, thereby strengthening its influence on generation. We therefore propose Structural Skeleton-guided Reasoning (SSR), which replaces suppression with structural decoupling by first generating an abstract functional skeleton and then using it to guide the full reasoning trace. Anchoring analyses show that SSR reduces all three forms of answer dependence. Across in-domain and out-of-distribution benchmarks, its distilled variant, SSR-D, improves performance by up to 10 points over suppression baselines and better preserves out-of-distribution performance. Code is available at this https URL.
Explanation based In-Context Demonstrations Retrieval for Multilingual Grammatical Error Correction
Grammatical error correction (GEC) aims to correct grammatical, spelling, and semantic errors in natural language text. With the growing of large language models (LLMs), direct text generation has gradually become the focus of the GEC methods, and few-shot in-context learning presents a cost-effective solution. However, selecting effective in-context examples remains challenging, as the similarity between input texts does not necessarily correspond to similar grammatical error patterns. In this paper, we propose a novel retrieval method based on natural language grammatical error explanations (GEE) to address this issue. Our method retrieves suitable few-shot demonstrations by matching the GEE of the test input with that of pre-constructed database samples, where explanations for erroneous samples are generated by LLMs. We conducted multilingual GEC few-shot experiments on both major open-source and closed-source LLMs. Experiments across five languages show that our method outperforms existing semantic and BM25-based retrieval techniques, without requiring additional training or language adaptation. This also suggests that matching error patterns is key to selecting examples.
Encode Errors: Representational Retrieval of In-Context Demonstrations for Multilingual Grammatical Error Correction
Grammatical Error Correction (GEC) involves detecting and correcting the wrong usage of grammar. While large language models (LLMs) with in-context learning (ICL) capabilities have shown significant progress on various natural language processing (NLP) tasks, their few-shot performance on GEC remains suboptimal. This is mainly due to the challenge of retrieving suitable in-context demonstrations that capture error patterns instead of semantic similarity. In this paper, we demonstrate that LLMs can inherently capture information related to grammatical errors through their internal states. From these states, we extract the Grammatical Error Representation (GER), an informative and semantically neutral encoding of grammatical errors. Our novel GER-based retrieval method significantly boosts performance in ICL settings on multilingual GEC datasets, improving the precision of correction. For high-resource languages, our results on 8B-sized open-source models match those of closed-source models such as Deepseek2.5 and GPT-4o-mini. For low-resource languages, our scores surpass the baseline by up to a factor of 1.20. This method provides a more precise and resource-efficient solution for multilingual GEC, offering a promising direction for interpretable GEC research.
Metrics
Citations per year
Coauthors
