Hanmeng Liu

Linyi Yang, Shuibai Zhang, Libo Qin et al. · pku

Pre-trained language models (PLMs) are known to improve the generalization performance of natural language understanding models by leveraging large amounts of data during the pre-training phase. However, the out-of-distribution (OOD) generalization problem remains a challenge in many NLP tasks, limiting the real-world deployment of these methods. This paper presents the first attempt at creating a unified benchmark named GLUE-X for evaluating OOD robustness in NLP models, highlighting the importance of OOD robustness and providing insights on how to measure the robustness of a model and how to improve it. The benchmark includes 13 publicly available datasets for OOD testing, and evaluations are conducted on 8 classic NLP tasks over 21 popularly used PLMs, including GPT-3 and GPT-3.5. Our findings confirm the need for improved OOD accuracy in NLP tasks, as significant performance degradation was observed in all settings compared to in-distribution (ID) accuracy.

Hanmeng Liu, Ruoxi Ning, Zhiyang Teng et al. · bytedance

Harnessing logical reasoning ability is a comprehensive natural language understanding endeavor. With the release of Generative Pretrained Transformer 4 (GPT-4), highlighted as "advanced" at reasoning tasks, we are eager to learn the GPT-4 performance on various logical reasoning tasks. This report analyses multiple logical reasoning datasets, with popular benchmarks like LogiQA and ReClor, and newly-released datasets like AR-LSAT. We test the multi-choice reading comprehension and natural language inference tasks with benchmarks requiring logical reasoning. We further construct a logical reasoning out-of-distribution dataset to investigate the robustness of ChatGPT and GPT-4. We also make a performance comparison between ChatGPT and GPT-4. Experiment results show that ChatGPT performs significantly better than the RoBERTa fine-tuning method on most logical reasoning benchmarks. With early access to the GPT-4 API we are able to conduct intense experiments on the GPT-4 model. The results show GPT-4 yields even higher performance on most logical reasoning datasets. Among benchmarks, ChatGPT and GPT-4 do relatively well on well-known datasets like LogiQA and ReClor. However, the performance drops significantly when handling newly released and out-of-distribution datasets. Logical reasoning remains challenging for ChatGPT and GPT-4, especially on out-of-distribution and natural language inference datasets. We release the prompt-style logical reasoning datasets as a benchmark suite and name it LogiEval.

Hanmeng liu, Zhiyang Teng, Ruoxi Ning et al. · bytedance

Large language models (LLMs) have shown significant general language understanding abilities. However, there has been a scarcity of attempts to assess the logical reasoning capacities of these LLMs, an essential facet of natural language understanding. To encourage further investigation in this area, we introduce GLoRE, a General Logical Reasoning Evaluation platform that not only consolidates diverse datasets but also standardizes them into a unified format suitable for evaluating large language models across zero-shot and few-shot scenarios. Our experimental results show that compared to the performance of humans and supervised fine-tuning models, the logical reasoning capabilities of large reasoning models, such as OpenAI's o1 mini, DeepSeek R1 and QwQ-32B, have seen remarkable improvements, with QwQ-32B achieving the highest benchmark performance to date. GLoRE is designed as a living project that continuously integrates new datasets and models, facilitating robust and comparative assessments of model performance in both commercial and Huggingface communities.

Zhizhang Fu, Yuancheng Gu, Chenkai Hu et al.

Test-time scaling has enabled Large Language Models (LLMs) to tackle complex reasoning, yet the limitations of current Chain-of-Thought (CoT) evaluation obscures whether performance gains stem from genuine reasoning or mere verbosity. To address this, (1) we propose a novel neuro-symbolic framework for the non-intrusive, comprehensive process-centric evaluation of reasoning. (2) Through this lens, we identify four distinct behavioral prototypes and diagnose the failure modes. (3) We examine the impact of inference mode, training strategy, and model scale. Our analysis reveals that extended token generation is not a prerequisite for deep reasoning. Furthermore, we reveal critical constraints: mixing long and short CoT data in training risks in premature saturation and collapse, while distillation into smaller models captures behavioral length but fails to replicate logical efficacy due to intrinsic capacity limits.

Yiwen Chen, Fuwei Zhang, Zehao Chen et al.

Large Language Models (LLMs) have demonstrated powerful reasoning capabilities through Chain-of-Thought (CoT) in various tasks, yet the inefficiency of token-by-token generation hinders real-world deployment in latency-sensitive recommender systems. Latent reasoning has emerged as an effective paradigm in LLMs, performing multi-step inference in a continuous hidden-state space to achieve stronger reasoning at lower cost. However, this paradigm remains underexplored in mainstream generative recommendation. Adapting it reveals three unique challenges: (1) the gap between prior-less Semantic ID (SID) symbols and continuous latent reasoning - SIDs lack pre-trained semantics, hindering joint optimization; (2) representation drift due to a lack of reasoning chain supervision; and (3) the suboptimality of applying a globally fixed reasoning depth. To address these, we propose LASAR (Latent Adaptive Semantic Aligned Reasoning), an SFT-then-RL framework. First, we bridge this gap via two-stage training: Stage 1 grounds SID semantics before Stage 2 introduces latent reasoning, ensuring efficient convergence. Second, we mitigate representation drift through explicit CoT semantic alignment. Step-wise bidirectional KL divergence constrains the latent reasoning trajectory using hidden-state anchors extracted from CoT text, while a Policy Head predicts per-sample reasoning depth. Third, during the GRPO-based RL phase, terminal-only KL alignment accommodates variable-length reasoning, and REINFORCE optimizes the Policy Head to dynamically allocate steps. This nearly halves the average latent step count while simultaneously improving recommendation quality. Experiments on three real-world datasets demonstrate that LASAR outperforms all baselines. It adds marginal inference latency and is roughly 20 times faster than generating explicit CoT text.

Mengru Ding, Hanmeng Liu, Zhizhang Fu et al.

Recent advancements in Chain-of-Thought (CoT) reasoning utilize complex modules but are hampered by high token consumption, limited applicability, and challenges in reproducibility. This paper conducts a critical evaluation of CoT prompting, extending beyond arithmetic to include complex logical and commonsense reasoning tasks, areas where standard CoT methods fall short. We propose the integration of human-like heuristics and shortcuts into language models (LMs) through "break the chain" strategies. These strategies disrupt traditional CoT processes using controlled variables to assess their efficacy. Additionally, we develop innovative zero-shot prompting strategies that encourage the use of shortcuts, enabling LMs to quickly exploit reasoning clues and bypass detailed procedural steps. Our comprehensive experiments across various LMs, both commercial and open-source, reveal that LMs maintain effective performance with "break the chain" strategies. We also introduce ShortcutQA, a dataset specifically designed to evaluate reasoning through shortcuts, compiled from competitive tests optimized for heuristic reasoning tasks such as forward/backward reasoning and simplification. Our analysis confirms that ShortcutQA not only poses a robust challenge to LMs but also serves as an essential benchmark for enhancing reasoning efficiency in AI.

Jian Liu, Leyang Cui, Hanmeng Liu et al.

Machine reading is a fundamental task for testing the capability of natural language understanding, which is closely related to human cognition in many aspects. With the rising of deep learning techniques, algorithmic models rival human performances on simple QA, and thus increasingly challenging machine reading datasets have been proposed. Though various challenges such as evidence integration and commonsense knowledge have been integrated, one of the fundamental capabilities in human reading, namely logical reasoning, is not fully investigated. We build a comprehensive dataset, named LogiQA, which is sourced from expert-written questions for testing human Logical reasoning. It consists of 8,678 QA instances, covering multiple types of deductive reasoning. Results show that state-of-the-art neural models perform by far worse than human ceiling. Our dataset can also serve as a benchmark for reinvestigating logical AI under the deep learning NLP setting. The dataset is freely available at https://github.com/lgw863/LogiQA-dataset

Hanmeng Liu, Shichao Weng, Xiulai Liu et al.

Multiple-choice reasoning benchmarks face dual challenges: rapid saturation from advancing models and data contamination that undermines static evaluations. Ad-hoc hardening methods (paraphrasing, perturbation) attempt to increase difficulty but sacrifice logical validity for surface complexity, falling short to challenge advanced reasoning models. We present LogiHard, a formal framework that deterministically transforms 0-order selection into 2-order logical judgment, which significantly increases the thinking overhead and reasoning steps. The framework integrates Item Response Theory (IRT) for computerized adaptive testing (CAT), enabling precise difficulty control with fewer questions than static benchmarks. We instantiate LogiHard-2k, a logical reasoning dataset constructed by cognitively ranking high-stakes examination questions via 9-dimensional analysis of model thinking traces, followed by combinatorial transformation of high-difficulty items. Evaluation across twelve state-of-the-art models reveals an accuracy degradation ranging from 31% to 56% on combinatorially hardened questions. LLMs suffer from the multi-select failure and early exit bias, which are not shared by human testees. Zero-shot transfer to MMLU demonstrates 47% accuracy degradation (89.84% to 42.86%), confirming applicability across domains with provable validity preservation. The consistent aggregate degeneration is domain-agnostic and stems not from knowledge deficits but from a combinatorial reasoning gap, reflecting a training-induced completeness-verification deficit.

Hanmeng Liu, Zhizhang Fu, Mengru Ding et al.

With the emergence of advanced reasoning models like OpenAI o3 and DeepSeek-R1, large language models (LLMs) have demonstrated remarkable reasoning capabilities. However, their ability to perform rigorous logical reasoning remains an open question. This survey synthesizes recent advancements in logical reasoning within LLMs, a critical area of AI research. It outlines the scope of logical reasoning in LLMs, its theoretical foundations, and the benchmarks used to evaluate reasoning proficiency. We analyze existing capabilities across different reasoning paradigms - deductive, inductive, abductive, and analogical - and assess strategies to enhance reasoning performance, including data-centric tuning, reinforcement learning, decoding strategies, and neuro-symbolic approaches. The review concludes with future directions, emphasizing the need for further exploration to strengthen logical reasoning in AI systems.

Hanmeng Liu, Zhiyang Teng, Chaoli Zhang et al. · bytedance

Chain-of-Thought (CoT) prompting has emerged as a pivotal technique for augmenting the inferential capabilities of language models during reasoning tasks. Despite its advancements, CoT often grapples with challenges in validating reasoning validity and ensuring informativeness. Addressing these limitations, this paper introduces the Logic Agent (LA), an agent-based framework aimed at enhancing the validity of reasoning processes in Large Language Models (LLMs) through strategic logic rule invocation. Unlike conventional approaches, LA transforms LLMs into logic agents that dynamically apply propositional logic rules, initiating the reasoning process by converting natural language inputs into structured logic forms. The logic agent leverages a comprehensive set of predefined functions to systematically navigate the reasoning process. This methodology not only promotes the structured and coherent generation of reasoning constructs but also significantly improves their interpretability and logical coherence. Through extensive experimentation, we demonstrate LA's capacity to scale effectively across various model sizes, markedly improving the precision of complex reasoning across diverse tasks.

Shichao Weng, Zhiqiang Wang, Yuhua Zhou et al.

Geometric Problem Solving (GPS) poses a unique challenge for Multimodal Large Language Models (MLLMs), requiring not only the joint interpretation of text and diagrams but also iterative visuospatial reasoning. While existing approaches process diagrams as static images, they lack the capacity for dynamic manipulation - a core aspect of human geometric reasoning involving auxiliary line construction and affine transformations. We present GeoSketch, a neural-symbolic framework that recasts geometric reasoning as an interactive perception-reasoning-action loop. GeoSketch integrates: (1) a Perception module that abstracts diagrams into structured logic forms, (2) a Symbolic Reasoning module that applies geometric theorems to decide the next deductive step, and (3) a Sketch Action module that executes operations such as drawing auxiliary lines or applying transformations, thereby updating the diagram in a closed loop. To train this agent, we develop a two-stage pipeline: supervised fine-tuning on 2,000 symbolic-curated trajectories followed by reinforcement learning with dense, symbolic rewards to enhance robustness and strategic exploration. To evaluate this paradigm, we introduce the GeoSketch Benchmark, a high-quality set of 390 geometry problems requiring auxiliary construction or affine transformations. Experiments on strong MLLM baselines demonstrate that GeoSketch significantly improves stepwise reasoning accuracy and problem-solving success over static perception methods. By unifying hierarchical decision-making, executable visual actions, and symbolic verification, GeoSketch advances multimodal reasoning from static interpretation to dynamic, verifiable interaction, establishing a new foundation for solving complex visuospatial problems.

Hanmeng Liu, Yiran Ding, Zhizhang Fu et al.

Large reasoning models, often post-trained on long chain-of-thought (long CoT) data with reinforcement learning, achieve state-of-the-art performance on mathematical, coding, and domain-specific reasoning benchmarks. However, their logical reasoning capabilities - fundamental to human cognition and independent of domain knowledge - remain understudied. To address this gap, we introduce LogiEval, a holistic benchmark for evaluating logical reasoning in large reasoning models. LogiEval spans diverse reasoning types (deductive, inductive, analogical, and abductive) and task formats (e.g., logical sequence, argument analysis), sourced from high-quality human examinations (e.g., LSAT, GMAT). Our experiments demonstrate that modern reasoning models excel at 4-choice argument analysis problems and analogical reasoning, surpassing human performance, yet exhibit uneven capabilities across reasoning types and formats, highlighting limitations in their generalization. Our analysis reveals that human performance does not mirror model failure distributions. To foster further research, we curate LogiEval-Hard, a challenging subset identified through a novel screening paradigm where small-model failures (Qwen3-30B-A3B) reliably predict difficulties for larger models. Modern models show striking, consistent failures on LogiEval-Hard. This demonstrates that fundamental reasoning bottlenecks persist across model scales, and establishes LogiEval-Hard as both a diagnostic tool and a rigorous testbed for advancing logical reasoning in LLMs.

Hanmeng Liu, Zhiyang Teng, Leyang Cui et al.

Generative Pre-trained Transformer 4 (GPT-4) demonstrates impressive chain-of-thought reasoning ability. Recent work on self-instruction tuning, such as Alpaca, has focused on enhancing the general proficiency of models. These instructions enable the model to achieve performance comparable to GPT-3.5 on general tasks like open-domain text generation and paraphrasing. However, they fall short of helping the model handle complex reasoning tasks. To bridge the gap, this paper presents LogiCoT, a new instruction-tuning dataset for Logical Chain-of-Thought reasoning with GPT-4. We elaborate on the process of harvesting instructions for prompting GPT-4 to generate chain-of-thought rationales. LogiCoT serves as an instruction set for teaching models of logical reasoning and elicits general reasoning skills.

Jian Liu, Zhiyang Teng, Leyang Cui et al.

Aspect category sentiment analysis has attracted increasing research attention. The dominant methods make use of pre-trained language models by learning effective aspect category-specific representations, and adding specific output layers to its pre-trained representation. We consider a more direct way of making use of pre-trained language models, by casting the ACSA tasks into natural language generation tasks, using natural language sentences to represent the output. Our method allows more direct use of pre-trained knowledge in seq2seq language models by directly following the task setting during pre-training. Experiments on several benchmarks show that our method gives the best reported results, having large advantages in few-shot and zero-shot settings.

Hanmeng Liu, Leyang Cui, Jian Liu et al.

Natural language inference (NLI) is a fundamental NLP task, investigating the entailment relationship between two texts. Popular NLI datasets present the task at sentence-level. While adequate for testing semantic representations, they fall short for testing contextual reasoning over long texts, which is a natural part of the human inference process. We introduce ConTRoL, a new dataset for ConTextual Reasoning over Long texts. Consisting of 8,325 expert-designed "context-hypothesis" pairs with gold labels, ConTRoL is a passage-level NLI dataset with a focus on complex contextual reasoning types such as logical reasoning. It is derived from competitive selection and recruitment test (verbal reasoning test) for police recruitment, with expert level quality. Compared with previous NLI benchmarks, the materials in ConTRoL are much more challenging, involving a range of reasoning types. Empirical results show that state-of-the-art language models perform by far worse than educated humans. Our dataset can also serve as a testing-set for downstream tasks like Checking Factual Correctness of Summaries.