Publications

2026

1. ICML

   

   How Language Models Process Negation

   Zhejian Zhou, Tianyi Zhou, Robin Jia, and Jonathan May

   In Forty-third International Conference on Machine Learning, 2026

   Abs

   We study how Large Language Models (LLMs) process negation mechanistically. First, we establish that even though open-weight models often provide wrong answers to questions involving negation, they do possess internal components that process negation correctly. Their poor accuracy is due to late-layer attention behavior that promotes simple shortcuts; ablating those attention modules greatly improves accuracy on negation-related questions. Second, we uncover how models process negation. We consider two hypotheses: models could use attention heads that attend to the phrase being negated and suppress related concepts, or they could directly construct a representation of the entire negative phrase (e.g., representing “not gas” as a vector that promotes liquids and solids). We apply a range of observational and causal interpretability techniques on Mistral-7B and Llama-3.1-8B to show that models implement both mechanisms, with the “constructive” mechanism being more prominent. Combined, our work deepens the understanding of LLMs’ internals, highlighting construction-dominant computations and the coexistence of competing mechanisms within LLMs.

2024

1. EMNLP Findings

   

   Scaling Behavior for Large Language Models regarding Numeral Systems: An Example using Pythia

   Zhejian Zhou, JIayu Wang, Dahua Lin, and Kai Chen

   In Findings of the Association for Computational Linguistics: EMNLP 2024, Nov 2024

   Abs

   Though Large Language Models (LLMs) have shown remarkable abilities in mathematics reasoning, they are still struggling with performing numeric operations accurately, such as addition and multiplication. Numbers can be tokenized into tokens in various ways by different LLMs and affect the numeric operations performance. Currently, there are two representatives: 1) Tokenize into 1-digit, and 2) Tokenize into 1∼3 digit. The difference is roughly equivalent to using different numeral systems (namely base 10 or base 10^3). In light of this, we study the scaling behavior of different numeral systems in the context of transformer-based large language models. We empirically show that a base 10 system is consistently more data-efficient than a base 10^2 or 10^3 system across training data scale, model sizes under from-scratch training settings, while different number systems have very similar fine-tuning performances. We attribute this to higher token frequencies of a base 10 system. Additionally, we reveal extrapolation behavior patterns on addition and multiplication. We identify that base 100 and base 1000 systems struggle on token-level discernment and token-level operations. We also sheds light on the mechanism learnt by the models.

2. arxiv

   

   InternLM2.5-StepProver: Advancing Automated Theorem Proving via Expert Iteration on Large-Scale LEAN Problems

   Zijian Wu, Suozhi Huang, Zhejian Zhou, Huaiyuan Ying, Jiayu Wang, Dahua Lin, and Kai Chen

   Nov 2024

   HTML
3. arxiv

   

   Internlm-math: Open math large language models toward verifiable reasoning

   Huaiyuan Ying, Shuo Zhang, Linyang Li, Zhejian Zhou, Yunfan Shao, Zhaoye Fei, Yichuan Ma, Jiawei Hong, Kuikun Liu, Ziyi Wang, and  others

   arXiv preprint arXiv:2402.06332, Nov 2024

   Abs HTML Code

   

   The math abilities of large language models can represent their abstract reasoning ability. In this paper, we introduce and open-source our math reasoning LLMs InternLM-Math which is continue pre-trained from InternLM2. We unify chain-of-thought reasoning, reward modeling, formal reasoning, data augmentation, and code interpreter in a unified seq2seq format and supervise our model to be a versatile math reasoner, verifier, prover, and augmenter. These abilities can be used to develop the next math LLMs or self-iteration. InternLM-Math obtains open-sourced state-of-the-art performance under the setting of in-context learning, supervised fine-tuning, and code-assisted reasoning in various informal and formal benchmarks including GSM8K, MATH, Hungary math exam, MathBench-ZH, and MiniF2F. Our pre-trained model achieves 30.3 on the MiniF2F test set without fine-tuning. We further explore how to use LEAN to solve math problems and study its performance under the setting of multi-task learning which shows the possibility of using LEAN as a unified platform for solving and proving in math. Our models, codes, and data are released at https://github.com/InternLM/InternLM-Math.

4. ICML Spotlight

   

   StackSight: Unveiling WebAssembly through Large Language Models and Neurosymbolic Chain-of-Thought Decompilation

   Weike Fang, Zhejian Zhou, Junzhou He, and Weihang Wang

   In Proceedings of the 41st International Conference on Machine Learning, 21–27 jul 2024

   Abs HTML PDF

   WebAssembly enables near-native execution in web applications and is increasingly adopted for tasks that demand high performance and robust security. However, its assembly-like syntax, implicit stack machine, and low-level data types make it extremely difficult for human developers to understand, spurring the need for effective WebAssembly reverse engineering techniques. In this paper, we propose StackSight, a novel neurosymbolic approach that combines Large Language Models (LLMs) with advanced program analysis to decompile complex WebAssembly code into readable C++ snippets. StackSight visualizes and tracks virtual stack alterations via a static analysis algorithm and then applies chain-of-thought prompting to harness LLM’s complex reasoning capabilities. Evaluation results show that StackSight significantly improves WebAssembly decompilation. Our user study also demonstrates that code snippets generated by StackSight have significantly higher win rates and enable a better grasp of code semantics.

5. MIR

   

   MOSS: An Open Conversational Large Language Model

   Tianxiang Sun, Xiaotian Zhang, Zhengfu He, Peng Li, Qinyuan Cheng, Xiangyang Liu, Hang Yan, Yunfan Shao, Qiong Tang, Shiduo Zhang, Xingjian Zhao, Ke Chen, Yining Zheng, Zhejian Zhou, Ruixiao Li, Jun Zhan, Yunhua Zhou, Linyang Li, Xiaogui Yang, Lingling Wu, Zhangyue Yin, Xuanjing Huang, Yu-Gang Jiang, and Xipeng Qiu

   Machine Intelligence Research, May 2024

   Abs HTML Code

   

   Conversational large language models (LLMs) such as ChatGPT and GPT-4 have recently exhibited remarkable capabilities across various domains, capturing widespread attention from the public. To facilitate this line of research, in this paper, we report the development of MOSS, an open-sourced conversational LLM that contains 16 B parameters and can perform a variety of instructions in multi-turn interactions with humans. The base model of MOSS is pre-trained on large-scale unlabeled English, Chinese, and code data. To optimize the model for dialogue, we generate 1.1 M synthetic conversations based on user prompts collected through our earlier versions of the model API. We then perform preference-aware training on preference data annotated from AI feedback. Evaluation results on real-world use cases and academic benchmarks demonstrate the effectiveness of the proposed approaches. In addition, we present an effective practice to augment MOSS with several external tools. Through the development of MOSS, we have established a complete technical roadmap for large language models from pre-training, supervised fine-tuning to alignment, verifying the feasibility of chatGPT under resource-limited conditions and providing a reference for both the academic and industrial communities. Model weights and code are publicly available at https://github.com/OpenMOSS/MOSS.

2022

1. COLING

   

   ConFiguRe: Exploring Discourse-level Chinese Figures of Speech

   Dawei Zhu, Qiusi Zhan, Zhejian Zhou, Yifan Song, Jiebin Zhang, and Sujian Li

   In Proceedings of the 29th International Conference on Computational Linguistics, Oct 2022

   Abs HTML Code

   Figures of speech, such as metaphor and irony, are ubiquitous in literature works and colloquial conversations. This poses great challenge for natural language understanding since figures of speech usually deviate from their ostensible meanings to express deeper semantic implications. Previous research lays emphasis on the literary aspect of figures and seldom provide a comprehensive exploration from a view of computational linguistics. In this paper, we first propose the concept of figurative unit, which is the carrier of a figure. Then we select 12 types of figures commonly used in Chinese, and build a Chinese corpus for Contextualized Figure Recognition (ConFiguRe). Different from previous token-level or sentence-level counterparts, ConFiguRe aims at extracting a figurative unit from discourse-level context, and classifying the figurative unit into the right figure type. On ConFiguRe, three tasks, i.e., figure extraction, figure type classification and figure recognition, are designed and the state-of-the-art techniques are utilized to implement the benchmarks. We conduct thorough experiments and show that all three tasks are challenging for existing models, thus requiring further research. Our dataset and code are publicly available at \urlhttps://github.com/pku-tangent/ConFiguRe.