Ling-I Wu

Yicheng Zou, Dongsheng Zhu, Lin Zhu et al.

We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, including chemistry, materials, life sciences, and earth sciences. Achieving this massive scale is made possible by the robust infrastructure support of XTuner and LMDeploy, which facilitates highly efficient Reinforcement Learning (RL) training at the 1-trillion parameter level while ensuring strict precision consistency between training and inference. By seamlessly integrating these advancements, Intern-S1-Pro further fortifies the fusion of general and specialized intelligence, working as a Specializable Generalist, demonstrating its position in the top tier of open-source models for general capabilities, while outperforming proprietary models in the depth of specialized scientific tasks.

Lei Bai, Zhongrui Cai, Yuhang Cao et al.

In recent years, a plethora of open-source foundation models have emerged, achieving remarkable progress in some widely attended fields, with performance being quite close to that of closed-source models. However, in high-value but more challenging scientific professional fields, either the fields still rely on expert models, or the progress of general foundation models lags significantly compared to those in popular areas, far from sufficient for transforming scientific research and leaving substantial gap between open-source models and closed-source models in these scientific domains. To mitigate this gap and explore a step further toward Artificial General Intelligence (AGI), we introduce Intern-S1, a specialized generalist equipped with general understanding and reasoning capabilities with expertise to analyze multiple science modal data. Intern-S1 is a multimodal Mixture-of-Experts (MoE) model with 28 billion activated parameters and 241 billion total parameters, continually pre-trained on 5T tokens, including over 2.5T tokens from scientific domains. In the post-training stage, Intern-S1 undergoes offline and then online reinforcement learning (RL) in InternBootCamp, where we propose Mixture-of-Rewards (MoR) to synergize the RL training on more than 1000 tasks simultaneously. Through integrated innovations in algorithms, data, and training systems, Intern-S1 achieved top-tier performance in online RL training. On comprehensive evaluation benchmarks, Intern-S1 demonstrates competitive performance on general reasoning tasks among open-source models and significantly outperforms open-source models in scientific domains, surpassing closed-source state-of-the-art models in professional tasks, such as molecular synthesis planning, reaction condition prediction, predicting thermodynamic stabilities for crystals. Our models are available at https://huggingface.co/internlm/Intern-S1.

Minyu Chen, Song Qin, Ling-I Wu et al.

LLM-based automatic heuristic design has shown promise for generating executable heuristics for combinatorial optimization, but existing methods mainly rely on delayed endpoint performance. We propose a \emph{teacher-aware evolutionary framework} that uses independently trained learned optimization policies as behavioral teachers. Instead of deploying or imitating the teacher, our method queries it on states visited by candidate heuristic programs and uses its action preferences as local feedback for evolution. The resulting search discovers static executable heuristics guided by both task performance and teacher-derived behavioral signals. Experiments on scheduling, routing, and graph optimization benchmarks show that our method improves over performance-driven LLM heuristic evolution baselines while requiring no neural inference at deployment. These results suggest that learned optimization policies can be repurposed as behavioral feedback sources for automatic heuristic discovery.

Ruibang Liu, Minyu Chen, Ling-I Wu et al.

Automated program verification has always been an important component of building trustworthy software. While the analysis of real-world programs remains a theoretical challenge, the automation of loop invariant analysis has effectively resolved the problem. However, real-world programs that often mix complex data structures and control flows pose challenges to traditional loop invariant generation tools. To enhance the applicability of invariant generation techniques, we proposed ACInv, an Automated Complex program loop Invariant generation tool, which combines static analysis with Large Language Models (LLMs) to generate the proper loop invariants. We utilize static analysis to extract the necessary information for each loop and embed it into prompts for the LLM to generate invariants for each loop. Subsequently, we employ an LLM-based evaluator to assess the generated invariants, refining them by either strengthening, weakening, or rejecting them based on their correctness, ultimately obtaining enhanced invariants. We conducted experiments on ACInv, which showed that ACInv outperformed previous tools on data sets with data structures, and maintained similar performance to the state-of-the-art tool AutoSpec on numerical programs without data structures. For the total data set, ACInv can solve 21% more examples than AutoSpec and can generate reference data structure templates.

Minyu Chen, Guoqiang Li, Ling-I Wu et al.

Transformer-based large language models (LLMs) have demonstrated significant potential in addressing logic problems. capitalizing on the great capabilities of LLMs for code-related activities, several frameworks leveraging logical solvers for logic reasoning have been proposed recently. While existing research predominantly focuses on viewing LLMs as natural language logic solvers or translators, their roles as logic code interpreters and executors have received limited attention. This study delves into a novel aspect, namely logic code simulation, which forces LLMs to emulate logical solvers in predicting the results of logical programs. To further investigate this novel task, we formulate our three research questions: Can LLMs efficiently simulate the outputs of logic codes? What strength arises along with logic code simulation? And what pitfalls? To address these inquiries, we curate three novel datasets tailored for the logic code simulation task and undertake thorough experiments to establish the baseline performance of LLMs in code simulation. Subsequently, we introduce a pioneering LLM-based code simulation technique, Dual Chains of Logic (DCoL). This technique advocates a dual-path thinking approach for LLMs, which has demonstrated state-of-the-art performance compared to other LLM prompt strategies, achieving a notable improvement in accuracy by 7.06% with GPT-4-Turbo.

Xu Guo, Tianyi Liang, Tong Jian et al.

Reinforcement Learning with Verifiable Rewards (RLVR) improves instruction following capabilities of large language models (LLMs), but suffers from training inefficiency due to inadequate difficulty assessment. Moreover, RLVR is prone to over-optimization, where LLMs exploit verification shortcuts without aligning to the actual intent of user instructions. We introduce Instruction Following Decorator (IFDecorator}, a framework that wraps RLVR training into a robust and sample-efficient pipeline. It consists of three components: (1) a cooperative-adversarial data flywheel that co-evolves instructions and hybrid verifications, generating progressively more challenging instruction-verification pairs; (2) IntentCheck, a bypass module enforcing intent alignment; and (3) trip wires, a diagnostic mechanism that detects reward hacking via trap instructions, which trigger and capture shortcut exploitation behaviors. Our Qwen2.5-32B-Instruct-IFDecorator achieves 87.43% accuracy on IFEval, outperforming larger proprietary models such as GPT-4o. Additionally, we demonstrate substantial improvements on FollowBench while preserving general capabilities. Our trip wires show significant reductions in reward hacking rates. We will release models, code, and data for future research.

Junyong Chen, Ling-I Wu, Minyu Chen et al.

Automated Rule Checking (ARC) plays a crucial role in advancing the construction industry by addressing the laborious, inconsistent, and error-prone nature of traditional model review conducted by industry professionals. Manual assessment against intricate sets of rules often leads to significant project delays and expenses. In response to these challenges, ARC offers a promising solution to improve efficiency and compliance in design within the construction sector. However, the main challenge of ARC lies in translating regulatory text into a format suitable for computer processing. Current methods for rule interpretation require extensive manual labor, thereby limiting their practicality. To address this issue, our study introduces a novel approach that decomposes ARC into two distinct tasks: rule information extraction and verification code generation. Leveraging generative pre-trained transformers, our method aims to streamline the interpretation of regulatory texts and simplify the process of generating model compliance checking code. Through empirical evaluation and case studies, we showcase the effectiveness and potential of our approach in automating code compliance checking, enhancing the efficiency and reliability of construction projects.