Code data in large language model (LLM) pretraining is recognized crucial not only for code-related tasks but also for enhancing general intelligence of LLMs. Current open-source LLMs often heavily rely on human effort to produce their code pretraining data, such as employing hand-crafted filtering rules tailored to individual programming languages, or using human-annotated data to train quality filters. However, these approaches are inherently limited in scalability, prone to subjective biases, and costly to extend and maintain across diverse programming languages. To address these challenges, we introduce Seed-Coder, a series of open-source LLMs comprising base, instruct and reasoning models of 8B size, minimizing human involvement in data construction. Our code pretraining data is produced by a model-centric data pipeline, which predominantly leverages LLMs for scoring and filtering code data. The instruct model is further trained via supervised fine-tuning and preference optimization, and the reasoning model leverages Long-Chain-of-Thought (LongCoT) reinforcement learning to improve multi-step code reasoning. Seed-Coder achieves state-of-the-art results among open-source models of similar size and even surpasses some much larger models, demonstrating superior performance in code generation, code completion, code editing, code reasoning, and software engineering tasks.
In recent years, code intelligence has gained increasing importance in the field of automated software engineering. Meanwhile, the widespread adoption of Pretrained Language Models (PLMs) and Large Language Models (LLMs) has raised concerns regarding data contamination and its potential impact on model performance evaluation. This paper presents a systematic empirical study to investigate the fine-grained data contamination on code intelligence tasks. Our study involves diverse representative PLMs, namely RoBERTa and GPT-2, and LLMs, namely LLaMA and StarCoder, covering three major tasks: code translation, code generation, and code summarization. We categorize contamination scenarios into four types according to the code intelligence practice, namely input-only, output-only, unpaired, and paired contamination settings, and construct corresponding experimental and control groups for exploration. Experimental results show that, under the pre-training, fine-tuning, and inference paradigm adopted by PLMs, even deliberately injecting paired contamination does not lead to significant performance overestimation. But direct inference or small-scale fine-tuning uncovers the contamination effects. In contrast, LLMs with pre-training and inference paradigm are significantly affected by the paired contamination. Apart from the above, other contamination scenarios have no impact on both PLMs and LLMs. Our findings challenge the conventional belief that contamination inevitably leads to performance overestimation, providing new insights into the evaluation and deployment of code intelligence models.
While large language models (LLMs) show promise in code generation, existing benchmarks neglect the flowchart-based code generation. To promote further research on flowchart-based code generation, this work presents Flow2Code, a novel benchmark for flowchart-based code generation evaluation. The evaluation dataset spans 15 programming languages and includes 5,622 code segments paired with 16,866 flowcharts of three types: code, UML, and pseudocode. Extensive experiments with 13 multimodal LLMs reveal that current LLMs can not generate code based on flowcharts perfectly. Besides, experiment results show that the supervised fine-tuning technique contributes greatly to the models' performance. We publicly release our code and datasets at https://github.com/hml-github/Flow2Code.
Smart contracts can implement and automate parts of legal contracts, but ensuring their legal compliance remains challenging. Existing approaches such as formal specification, verification, and model-based development require expertise in both legal and software development domains, as well as extensive manual effort. Given the recent advances of Large Language Models (LLMs) in code generation, we investigate their ability to generate legally compliant smart contracts directly from natural language legal contracts, addressing these challenges. We propose a novel suite of metrics to quantify legal compliance based on modeling both legal and smart contracts as processes and comparing their behaviors. We select four LLMs, generate 20 smart contracts based on five legal contracts, and analyze their legal compliance. We find that while all LLMs generate syntactically correct code, there is significant variance in their legal compliance with larger models generally showing higher levels of compliance. We also evaluate the proposed metrics against properties of software metrics, showing they provide fine-grained distinctions, enable nuanced comparisons, and are applicable across domains for code from any source, LLM or developer. Our results suggest that LLMs can assist in generating starter code for legally compliant smart contracts with strict reviews, and the proposed metrics provide a foundation for automated and self-improving development workflows.
LLMs have been extensively used for the task of automated code generation. In this work, we examine the applicability of LLMs for the related but relatively unexplored task of code-equivalence checking, i.e., given two programs, whether they are functionally equivalent or not. This is an important problem since benchmarking code equivalence can play a critical role in evaluating LLM capabilities for tasks such as code re-writing and code translation. Towards this end, we present CETBench - Code Equivalence with Transformations Benchmark, constructed via a repository of programs, where two programs in the repository may be solving the same or different tasks. Each instance in our dataset is obtained by taking a pair of programs in the repository and applying a random series of pre-defined code transformations, resulting in (non-)equivalent pairs. Our analysis on this dataset reveals a surprising finding that very simple code transformations in the underlying pair of programs can result in a significant drop in performance of SOTA LLMs for the task of code-equivalence checking. To remedy this, we present a simple fine-tuning-based approach to boost LLM performance on the transformed pairs of programs. Our approach for dataset generation is generic, and can be used with repositories with varying program difficulty levels and allows for applying varying numbers as well as kinds of transformations. In our experiments, we perform ablations over the difficulty level of original programs, as well as the kind of transformations used in generating pairs for equivalence checking. Our analysis presents deep insights into the working of LLMs for the task of code-equivalence, and points to the fact that they may still be far from what could be termed as a semantic understanding of the underlying code.
Advancements in generative models have enabled multi-agent systems (MAS) to perform complex virtual tasks such as writing and code generation, which do not generalize well to physical multi-agent robotic teams. Current frameworks often treat agents as conceptual task executors rather than physically embodied entities, and overlook critical real-world constraints such as spatial context, robotic capabilities (e.g., sensing and navigation). To probe this gap, we reconfigure and stress-test a hierarchical multi-agent robotic team built on the CrewAI framework in a simulated emergency department onboarding scenario. We identify five persistent failure modes: role misalignment; tool access violations; lack of in-time handling of failure reports; noncompliance with prescribed workflows; bypassing or false reporting of task completion. Based on this analysis, we propose three design guidelines emphasizing process transparency, proactive failure recovery, and contextual grounding. Our work informs the development of more resilient and robust multi-agent robotic systems (MARS), including opportunities to extend virtual multi-agent frameworks to the real world.
Large Language Model (LLM) based multi-agent systems have shown remarkable performance in various tasks, especially when enhanced through collaborative communication. However, current methods often rely on a fixed number of agents and static communication structures, limiting their ability to adapt to varying task complexities. In this paper, we propose Adaptive Graph Pruning (AGP), a novel task-adaptive multi-agent collaboration framework that jointly optimizes agent quantity (hard-pruning) and communication topology (soft-pruning). Specifically, our method employs a two-stage training strategy: firstly, independently training soft-pruning networks for different agent quantities to determine optimal agent-quantity-specific complete graphs and positional masks across specific tasks; and then jointly optimizing hard-pruning and soft-pruning within a maximum complete graph to dynamically configure the number of agents and their communication topologies per task. Extensive experiments demonstrate that our approach is: (1) High-performing, achieving state-of-the-art results across six benchmarks and consistently generalizes across multiple mainstream LLM architectures, with a increase in performance of $2.58\%\sim 9.84\%$; (2) Task-adaptive, dynamically constructing optimized communication topologies tailored to specific tasks, with an extremely high performance in all three task categories (general reasoning, mathematical reasoning, and code generation); (3) Token-economical, having fewer training steps and token consumption at the same time, with a decrease in token consumption of $90\%+$; and (4) Training-efficient, achieving high performance with very few training steps compared with other methods. The performance will surpass the existing baselines after about ten steps of training under six benchmarks.
Large Language Models (LLMs) offer transformative capabilities for hardware design automation, particularly in Verilog code generation. However, they also pose significant data security challenges, including Verilog evaluation data contamination, intellectual property (IP) design leakage, and the risk of malicious Verilog generation. We introduce SALAD, a comprehensive assessment that leverages machine unlearning to mitigate these threats. Our approach enables the selective removal of contaminated benchmarks, sensitive IP and design artifacts, or malicious code patterns from pre-trained LLMs, all without requiring full retraining. Through detailed case studies, we demonstrate how machine unlearning techniques effectively reduce data security risks in LLM-aided hardware design.
Large language models (LLMs) are playing an increasingly large role in domains such as code generation, including hardware code generation, where Verilog is the key language. However, the amount of publicly available Verilog code pales in comparison to the amount of code available for software languages like Python. In this work, we present hdl2v ("HDL-to-Verilog"), a dataset which seeks to increase the amount of available human-written Verilog data by translating or compiling three other hardware description languages - VHDL, Chisel, and PyMTL3 - to Verilog. Furthermore, we demonstrate the value of hdl2v in enhancing LLM Verilog generation by improving performance of a 32 billion-parameter open-weight model by up to 23% (pass@10) in VerilogEvalV2, without utilizing any data augmentation or knowledge distillation from larger models. We also show hdl2v's ability to boost the performance of a data augmentation-based fine-tuning approach by 63%. Finally, we characterize and analyze our dataset to better understand which characteristics of HDL-to-Verilog datasets can be expanded upon in future work for even better performance.
We present SwingArena, a competitive evaluation framework for Large Language Models (LLMs) that closely mirrors real-world software development workflows. Unlike traditional static benchmarks, SwingArena models the collaborative process of software iteration by pairing LLMs as submitters, who generate patches, and reviewers, who create test cases and verify the patches through continuous integration (CI) pipelines. To support these interactive evaluations, we introduce a retrieval-augmented code generation (RACG) module that efficiently handles long-context challenges by providing syntactically and semantically relevant code snippets from large codebases, supporting multiple programming languages (C++, Python, Rust, and Go). This enables the framework to scale across diverse tasks and contexts while respecting token limitations. Our experiments, using over 400 high-quality real-world GitHub issues selected from a pool of 2,300 issues, show that models like GPT-4o excel at aggressive patch generation, whereas DeepSeek and Gemini prioritize correctness in CI validation. SwingArena presents a scalable and extensible methodology for evaluating LLMs in realistic, CI-driven software development settings. More details are available on our project page: swing-bench.github.io
Cross-modal entity linking refers to the ability to align entities and their attributes across different modalities. While cross-modal entity linking is a fundamental skill needed for real-world applications such as multimodal code generation, fake news detection, or scene understanding, it has not been thoroughly studied in the literature. In this paper, we introduce a new task and benchmark to address this gap. Our benchmark, MATE, consists of 5.5k evaluation instances featuring visual scenes aligned with their textual representations. To evaluate cross-modal entity linking performance, we design a question-answering task that involves retrieving one attribute of an object in one modality based on a unique attribute of that object in another modality. We evaluate state-of-the-art Vision-Language Models (VLMs) and humans on this task, and find that VLMs struggle significantly compared to humans, particularly as the number of objects in the scene increases. Our analysis also shows that, while chain-of-thought prompting can improve VLM performance, models remain far from achieving human-level proficiency. These findings highlight the need for further research in cross-modal entity linking and show that MATE is a strong benchmark to support that progress.
Model distillation has become essential for creating smaller, deployable language models that retain larger system capabilities. However, widespread deployment raises concerns about resilience to adversarial manipulation. This paper investigates vulnerability of distilled models to adversarial injection of biased content during training. We demonstrate that adversaries can inject subtle biases into teacher models through minimal data poisoning, which propagates to student models and becomes significantly amplified. We propose two propagation modes: Untargeted Propagation, where bias affects multiple tasks, and Targeted Propagation, focusing on specific tasks while maintaining normal behavior elsewhere. With only 25 poisoned samples (0.25% poisoning rate), student models generate biased responses 76.9% of the time in targeted scenarios - higher than 69.4% in teacher models. For untargeted propagation, adversarial bias appears 6x-29x more frequently in student models on unseen tasks. We validate findings across six bias types (targeted advertisements, phishing links, narrative manipulations, insecure coding practices), various distillation methods, and different modalities spanning text and code generation. Our evaluation reveals shortcomings in current defenses - perplexity filtering, bias detection systems, and LLM-based autorater frameworks - against these attacks. Results expose significant security vulnerabilities in distilled models, highlighting need for specialized safeguards. We propose practical design principles for building effective adversarial bias mitigation strategies.
We introduce POLLUX, a comprehensive open-source benchmark designed to evaluate the generative capabilities of large language models (LLMs) in Russian. Our main contribution is a novel evaluation methodology that enhances the interpretability of LLM assessment. For each task type, we define a set of detailed criteria and develop a scoring protocol where models evaluate responses and provide justifications for their ratings. This enables transparent, criteria-driven evaluation beyond traditional resource-consuming, side-by-side human comparisons. POLLUX includes a detailed, fine-grained taxonomy of 35 task types covering diverse generative domains such as code generation, creative writing, and practical assistant use cases, totaling 2,100 manually crafted and professionally authored prompts. Each task is categorized by difficulty (easy/medium/hard), with experts constructing the dataset entirely from scratch. We also release a family of LLM-as-a-Judge (7B and 32B) evaluators trained for nuanced assessment of generative outputs. This approach provides scalable, interpretable evaluation and annotation tools for model development, effectively replacing costly and less precise human judgments.
Although multi-agent systems based on large language models show strong capabilities on multiple tasks, they are still limited by high computational overhead, information loss, and robustness. Inspired by ResNet's residual learning, we propose Residual Mixture-of-Agents (RMoA), integrating residual connections to optimize efficiency and reliability. To maximize information utilization from model responses while minimizing computational costs, we innovatively design an embedding-based diversity selection mechanism that greedily selects responses via vector similarity. Furthermore, to mitigate iterative information degradation, we introduce a Residual Extraction Agent to preserve cross-layer incremental information by capturing inter-layer response differences, coupled with a Residual Aggregation Agent for hierarchical information integration. Additionally, we propose an adaptive termination mechanism that dynamically halts processing based on residual convergence, further improving inference efficiency. RMoA achieves state-of-the-art performance on the benchmarks of across alignment, mathematical reasoning, code generation, and multitasking understanding, while significantly reducing computational overhead. Code is available at https://github.com/mindhunter01/RMoA.
In recent years, large language models (LLMs) have showcased significant advancements in code generation. However, most evaluation benchmarks are primarily oriented towards Python, making it difficult to evaluate other programming languages, such as Swift, with high quality. By examining widely established multilingual benchmarks like HumanEval-XL and MultiPL-E, we identified critical issues specific to their Swift components, making them insufficient or even irrelevant for assessing LLM coding capabilities on Swift. Unlike these existing approaches, which prioritize rapid scaling and generalization by automatically translating Python-centric benchmarks with LLMs, we adopt a quality-over-quantity methodology. We present SwiftEval, the first Swift-oriented benchmark consisting of 28 carefully hand-crafted problems, and evaluate 44 popular Code LLMs on it. Our results show significant LLM scores drop for problems requiring language-specific features, most noticeable in the models of smaller sizes.
Frequent cyber-attacks have elevated WebShell exploitation and defense to a critical research focus within network security. However, there remains a significant shortage of publicly available, well-categorized malicious-code datasets organized by obfuscation method. Existing malicious-code generation methods, which primarily rely on prompt engineering, often suffer from limited diversity and high redundancy in the payloads they produce. To address these limitations, we propose \textbf{RAWG}, a \textbf{R}eward-driven \textbf{A}utomated \textbf{W}ebshell Malicious-code \textbf{G}enerator designed for red-teaming applications. Our approach begins by categorizing webshell samples from common datasets into seven distinct types of obfuscation. We then employ a large language model (LLM) to extract and normalize key tokens from each sample, creating a standardized, high-quality corpus. Using this curated dataset, we perform supervised fine-tuning (SFT) on an open-source large model to enable the generation of diverse, highly obfuscated webshell malicious payloads. To further enhance generation quality, we apply Proximal Policy Optimization (PPO), treating malicious-code samples as "chosen" data and benign code as "rejected" data during reinforcement learning. Extensive experiments demonstrate that RAWG significantly outperforms current state-of-the-art methods in both payload diversity and escape effectiveness.
Automatic code generation has gained significant momentum with the advent of Large Language Models (LLMs) such as GPT-4. Although many studies focus on improving the effectiveness of LLMs for code generation, very limited work tries to understand the generated code's characteristics and leverage that to improve failed cases. In this paper, as the most straightforward characteristic of code, we investigate the relationship between code complexity and the success of LLM generated code. Using a large set of standard complexity metrics, we first conduct an empirical analysis to explore their correlation with LLM's performance on code generation (i.e., Pass@1). Using logistic regression models, we identify which complexity metrics are most predictive of code correctness. Building on these findings, we propose an iterative feedback method, where LLMs are prompted to generate correct code based on complexity metrics from previous failed outputs. We validate our approach across multiple benchmarks (i.e., HumanEval, MBPP, LeetCode, and BigCodeBench) and various LLMs (i.e., GPT-4o, GPT-3.5 Turbo, Llama 3.1, and GPT-o3 mini), comparing the results with two baseline methods: (a) zero-shot generation, and (b) iterative execution-based feedback without our code complexity insights. Experiment results show that our approach makes notable improvements, particularly with a smaller LLM (GPT3.5 Turbo), where, e.g., Pass@1 increased by 35.71% compared to the baseline's improvement of 12.5% on the HumanEval dataset. The study expands experiments to BigCodeBench and integrates the method with the Reflexion code generation agent, leading to Pass@1 improvements of 20% (GPT-4o) and 23.07% (GPT-o3 mini). The results highlight that complexity-aware feedback enhances both direct LLM prompting and agent-based workflows.
Large Language Models (LLMs) with reasoning are trained to iteratively generate and refine their answers before finalizing them, which can help with applications to mathematics and code generation. We apply code generation with reasoning LLMs to a specific task in the mathematical field of combinatorial design. This field studies diverse types of combinatorial designs, many of which have lists of open instances for which existence has not yet been determined. The Constructive Protocol CPro1 uses LLMs to generate search heuristics that have the potential to construct solutions to small open instances. Starting with a textual definition and a validity verifier for a particular type of design, CPro1 guides LLMs to select and implement strategies, while providing automated hyperparameter tuning and execution feedback. CPro1 with reasoning LLMs successfully solves long-standing open instances for 7 of 16 combinatorial design problems selected from the 2006 Handbook of Combinatorial Designs, including new solved instances for 3 of these (Bhaskar Rao Designs, Symmetric Weighing Matrices, Balanced Ternary Designs) that were unsolved by CPro1 with non-reasoning LLMs. It also solves open instances for several problems from recent (2025) literature, generating new Covering Sequences, Johnson Clique Covers, Deletion Codes, and a Uniform Nested Steiner Quadruple System.
With the increasing size of frontier LLMs, post-training quantization has become the standard for memory-efficient deployment. Recent work has shown that basic rounding-based quantization schemes pose security risks, as they can be exploited to inject malicious behaviors into quantized models that remain hidden in full precision. However, existing attacks cannot be applied to more complex quantization methods, such as the GGUF family used in the popular ollama and llama$.$cpp frameworks. In this work, we address this gap by introducing the first attack on GGUF. Our key insight is that the quantization error -- the difference between the full-precision weights and their (de-)quantized version -- provides sufficient flexibility to construct malicious quantized models that appear benign in full precision. Leveraging this, we develop an attack that trains the target malicious LLM while constraining its weights based on quantization errors. We demonstrate the effectiveness of our attack on three popular LLMs across nine GGUF quantization data types on three diverse attack scenarios: insecure code generation ($\Delta$=$88.7\%$), targeted content injection ($\Delta$=$85.0\%$), and benign instruction refusal ($\Delta$=$30.1\%$). Our attack highlights that (1) the most widely used post-training quantization method is susceptible to adversarial interferences, and (2) the complexity of quantization schemes alone is insufficient as a defense.
Traditional code instruction data synthesis methods suffer from limited diversity and poor logic. We introduce Infinite-Instruct, an automated framework for synthesizing high-quality question-answer pairs, designed to enhance the code generation capabilities of large language models (LLMs). The framework focuses on improving the internal logic of synthesized problems and the quality of synthesized code. First, "Reverse Construction" transforms code snippets into diverse programming problems. Then, through "Backfeeding Construction," keywords in programming problems are structured into a knowledge graph to reconstruct them into programming problems with stronger internal logic. Finally, a cross-lingual static code analysis pipeline filters invalid samples to ensure data quality. Experiments show that on mainstream code generation benchmarks, our fine-tuned models achieve an average performance improvement of 21.70% on 7B-parameter models and 36.95% on 32B-parameter models. Using less than one-tenth of the instruction fine-tuning data, we achieved performance comparable to the Qwen-2.5-Coder-Instruct. Infinite-Instruct provides a scalable solution for LLM training in programming. We open-source the datasets used in the experiments, including both unfiltered versions and filtered versions via static analysis. The data are available at https://github.com/xingwenjing417/Infinite-Instruct-dataset