Reproducibility and reliability remain pressing challenges for generative AI systems whose behavior can drift with each model update or prompt revision. We introduce GPR-bench, a lightweight, extensible benchmark that operationalizes regression testing for general purpose use cases. GPR-bench couples an open, bilingual (English and Japanese) dataset covering eight task categories (e.g., text generation, code generation, and information retrieval) and 10 scenarios in each task categories (80 total test cases for each language) with an automated evaluation pipeline that employs "LLM-as-a-Judge" scoring of correctness and conciseness. Experiments across three recent model versions - gpt-4o-mini, o3-mini, and o4-mini - and two prompt configurations (default versus concise-writing instruction) reveal heterogeneous quality. Our results show that newer models generally improve correctness, but the differences are modest and not statistically significant, suggesting that GPR-bench may not be sufficiently challenging to differentiate between recent model versions. In contrast, the concise-writing instruction significantly enhances conciseness (+12.37 pp, Mann-Whitney U test: p < 0.001, effect size r = 0.2995) with minimal degradations on accuracy (-1.7 pp), demonstrating the effectiveness of prompt engineering. Released under the MIT License, GPR- bench lowers the barrier to initiating reproducibility monitoring and provides a foundation for community-driven extensions, while also raising important considerations about benchmark design for rapidly evolving language models.
This paper introduces SecRepoBench, a benchmark to evaluate LLMs on secure code generation in real-world repositories. SecRepoBench has 318 code generation tasks in 27 C/C++ repositories, covering 15 CWEs. We evaluate 19 state-of-the-art LLMs using our benchmark and find that the models struggle with generating correct and secure code. In addition, the performance of LLMs to generate self-contained programs as measured by prior benchmarks do not translate to comparative performance at generating secure and correct code at the repository level in SecRepoBench. We show that the state-of-the-art prompt engineering techniques become less effective when applied to the repository level secure code generation problem. We conduct extensive experiments, including an agentic technique to generate secure code, to demonstrate that our benchmark is currently the most difficult secure coding benchmark, compared to previous state-of-the-art benchmarks. Finally, our comprehensive analysis provides insights into potential directions for enhancing the ability of LLMs to generate correct and secure code in real-world repositories.
Reinforcement Learning (RL) has demonstrated excellent decision-making potential in platoon coordination problems. However, due to the variability of coordination goals, the complexity of the decision problem, and the time-consumption of trial-and-error in manual design, finding a well performance reward function to guide RL training to solve complex platoon coordination problems remains challenging. In this paper, we formally define the Platoon Coordination Reward Design Problem (PCRDP), extending the RL-based cooperative platoon coordination problem to incorporate automated reward function generation. To address PCRDP, we propose a Large Language Model (LLM)-based Platoon coordination Reward Design (PCRD) framework, which systematically automates reward function discovery through LLM-driven initialization and iterative optimization. In this method, LLM first initializes reward functions based on environment code and task requirements with an Analysis and Initial Reward (AIR) module, and then iteratively optimizes them based on training feedback with an evolutionary module. The AIR module guides LLM to deepen their understanding of code and tasks through a chain of thought, effectively mitigating hallucination risks in code generation. The evolutionary module fine-tunes and reconstructs the reward function, achieving a balance between exploration diversity and convergence stability for training. To validate our approach, we establish six challenging coordination scenarios with varying complexity levels within the Yangtze River Delta transportation network simulation. Comparative experimental results demonstrate that RL agents utilizing PCRD-generated reward functions consistently outperform human-engineered reward functions, achieving an average of 10\% higher performance metrics in all scenarios.
Reinforcement learning (RL) has increasingly become a pivotal technique in the post-training of large language models (LLMs). The effective exploration of the output space is essential for the success of RL. We observe that for complex problems, during the early stages of training, the model exhibits strong exploratory capabilities and can identify promising solution ideas. However, its limited capability at this stage prevents it from successfully solving these problems. The early suppression of these potentially valuable solution ideas by the policy gradient hinders the model's ability to revisit and re-explore these ideas later. Consequently, although the LLM's capabilities improve in the later stages of training, it still struggles to effectively address these complex problems. To address this exploration issue, we propose a novel algorithm named Retrospective Replay-based Reinforcement Learning (RRL), which introduces a dynamic replay mechanism throughout the training process. RRL enables the model to revisit promising states identified in the early stages, thereby improving its efficiency and effectiveness in exploration. To evaluate the effectiveness of RRL, we conduct extensive experiments on complex reasoning tasks, including mathematical reasoning and code generation, and general dialogue tasks. The results indicate that RRL maintains high exploration efficiency throughout the training period, significantly enhancing the effectiveness of RL in optimizing LLMs for complicated reasoning tasks. Moreover, it also improves the performance of RLHF, making the model both safer and more helpful.
We introduce LeetCodeDataset, a high-quality benchmark for evaluating and training code-generation models, addressing two key challenges in LLM research: the lack of reasoning-focused coding benchmarks and self-contained training testbeds. By curating LeetCode Python problems with rich metadata, broad coverage, 100+ test cases per problem, and temporal splits (pre/post July 2024), our dataset enables contamination-free evaluation and efficient supervised fine-tuning (SFT). Experiments show reasoning models significantly outperform non-reasoning counterparts, while SFT with only 2.6K model-generated solutions achieves performance comparable to 110K-sample counterparts. The dataset and evaluation framework are available on Hugging Face and Github.
Scaling test-time computation, or affording a generator large language model (LLM) extra compute during inference, typically employs the help of external non-generative evaluators (i.e., reward models). Concurrently, LLM-judges, models trained to generate evaluations and critiques (explanations) in natural language, are becoming increasingly popular in automatic evaluation. Despite judge empirical successes, their effectiveness as evaluators in test-time scaling settings is largely unknown. In this paper, we introduce the Judge Evaluation for Test-Time Scaling (JETTS) benchmark, which evaluates judge performance in three domains (math reasoning, code generation, and instruction following) under three task settings: response reranking, step-level beam search, and critique-based response refinement. We evaluate 10 different judge models (7B-70B parameters) for 8 different base generator models (6.7B-72B parameters). Our benchmark shows that while judges are competitive with outcome reward models in reranking, they are consistently worse than process reward models in beam search procedures. Furthermore, though unique to LLM-judges, their natural language critiques are currently ineffective in guiding the generator towards better responses.
Large Language Models (LLMs) have advanced Verilog code generation significantly, yet face challenges in data quality, reasoning capabilities, and computational efficiency. This paper presents ReasoningV, a novel model employing a hybrid reasoning strategy that integrates trained intrinsic capabilities with dynamic inference adaptation for Verilog code generation. Our framework introduces three complementary innovations: (1) ReasoningV-5K, a high-quality dataset of 5,000 functionally verified instances with reasoning paths created through multi-dimensional filtering of PyraNet samples; (2) a two-stage training approach combining parameter-efficient fine-tuning for foundational knowledge with full-parameter optimization for enhanced reasoning; and (3) an adaptive reasoning mechanism that dynamically adjusts reasoning depth based on problem complexity, reducing token consumption by up to 75\% while preserving performance. Experimental results demonstrate ReasoningV's effectiveness with a pass@1 accuracy of 57.8\% on VerilogEval-human, achieving performance competitive with leading commercial models like Gemini-2.0-flash (59.5\%) and exceeding the previous best open-source model by 10.4 percentage points. ReasoningV offers a more reliable and accessible pathway for advancing AI-driven hardware design automation, with our model, data, and code available at https://github.com/BUAA-CLab/ReasoningV.
While large language models (LLMs) have been widely applied to code generation, they struggle with generating entire deep learning projects, which are characterized by complex structures, longer functions, and stronger reliance on domain knowledge than general-purpose code. An open-domain LLM often lacks coherent contextual guidance and domain expertise for specific projects, making it challenging to produce complete code that fully meets user requirements. In this paper, we propose a novel planning-guided code generation method, DLCodeGen, tailored for generating deep learning projects. DLCodeGen predicts a structured solution plan, offering global guidance for LLMs to generate the project. The generated plan is then leveraged to retrieve semantically analogous code samples and subsequently abstract a code template. To effectively integrate these multiple retrieval-augmented techniques, a comparative learning mechanism is designed to generate the final code. We validate the effectiveness of our approach on a dataset we build for deep learning code generation. Experimental results demonstrate that DLCodeGen outperforms other baselines, achieving improvements of 9.7% in CodeBLEU and 3.6% in human evaluation metrics.
Large Language Models (LLMs), such as GitHub Copilot and ChatGPT have become popular among programming students. Students use LLMs to assist them in programming courses, including generating source code. Previous work has evaluated the ability of LLMs in solving introductory-course programming assignments. The results have shown that LLMs are highly effective in generating code for introductory Computer Science (CS) courses. However, there is a gap in research on evaluating LLMs' ability to generate code that solves advanced programming assignments. In this work, we evaluate the ability of four LLM tools to solve programming assignments from advanced CS courses in three popular programming languages, Java, Python, and C. We manually select 12 problems, three problems from introductory courses as the baseline and nine programming assignments from second- and third-year CS courses. To evaluate the LLM-generated code, we generate a test suite of 1000 test cases per problem and analyze the program output. Our evaluation shows that although LLMs are highly effective in generating source code for introductory programming courses, solving advanced programming assignments is more challenging. Nonetheless, in many cases, LLMs identify the base problem and provide partial solutions that may be useful to CS students. Furthermore, our results may provide useful guidance for teachers of advanced programming courses on how to design programming assignments.
Large language models (LLMs) have transformed code generation, yet their application in hardware design produces gate counts 38\%--1075\% higher than human designs. We present CircuitMind, a multi-agent framework that achieves human-competitive efficiency through three key innovations: syntax locking (constraining generation to basic logic gates), retrieval-augmented generation (enabling knowledge-driven design), and dual-reward optimization (balancing correctness with efficiency). To evaluate our approach, we introduce TC-Bench, the first gate-level benchmark harnessing collective intelligence from the TuringComplete ecosystem -- a competitive circuit design platform with hundreds of thousands of players. Experiments show CircuitMind enables 55.6\% of model implementations to match or exceed top-tier human experts in composite efficiency metrics. Most remarkably, our framework elevates the 14B Phi-4 model to outperform both GPT-4o mini and Gemini 2.0 Flash, achieving efficiency comparable to the top 25\% of human experts without requiring specialized training. These innovations establish a new paradigm for hardware optimization where collaborative AI systems leverage collective human expertise to achieve optimal circuit designs. Our model, data, and code are open-source at https://github.com/BUAA-CLab/CircuitMind.
This tutorial on adaptation of LLMs is designed to address the growing demand for models that go beyond the static capabilities of generic LLMs by providing an overview of dynamic, domain-specific, and task-adaptive LLM adaptation techniques. While general LLMs have demonstrated strong generalization across a variety of tasks, they often struggle to perform well in specialized domains such as finance, healthcare, and code generation for underrepresented languages. Additionally, their static nature limits their ability to evolve with the changing world, and they are often extremely large in size, making them impractical and costly to deploy at scale. As a result, the adaptation of LLMs has drawn much attention since the birth of LLMs and is of core importance, both for industry, which focuses on serving its targeted users, and academia, which can greatly benefit from small but powerful LLMs. To address this gap, this tutorial aims to provide an overview of the LLM adaptation techniques. We start with an introduction to LLM adaptation, from both the data perspective and the model perspective. We then emphasize how the evaluation metrics and benchmarks are different from other techniques. After establishing the problems, we explore various adaptation techniques. We categorize adaptation techniques into two main families. The first is parametric knowledge adaptation, which focuses on updating the parametric knowledge within LLMs. Additionally, we will discuss real-time adaptation techniques, including model editing, which allows LLMs to be updated dynamically in production environments. The second kind of adaptation is semi-parametric knowledge adaptation, where the goal is to update LLM parameters to better leverage external knowledge or tools through techniques like retrieval-augmented generation (RAG) and agent-based systems.
Large Language Models (LLMs) are likely to play a key role in Intent-Based Networking (IBN) as they show remarkable performance in interpreting human language as well as code generation, enabling the translation of high-level intents expressed by humans into low-level network configurations. In this paper, we leverage closed-source language models (i.e., Google Gemini 1.5 pro, ChatGPT-4) and open-source models (i.e., LLama, Mistral) to investigate their capacity to generate E2E network configurations for radio access networks (RANs) and core networks in 5G/6G mobile networks. We introduce a novel performance metrics, known as FEACI, to quantitatively assess the format (F), explainability (E), accuracy (A), cost (C), and inference time (I) of the generated answer; existing general metrics are unable to capture these features. The results of our study demonstrate that open-source models can achieve comparable or even superior translation performance compared with the closed-source models requiring costly hardware setup and not accessible to all users.
Large Language Models (LLMs) have rapidly transformed software development, especially in code generation. However, their inconsistent performance, prone to hallucinations and quality issues, complicates program comprehension and hinders maintainability. Research indicates that prompt engineering-the practice of designing inputs to direct LLMs toward generating relevant outputs-may help address these challenges. In this regard, researchers have introduced prompt patterns, structured templates intended to guide users in formulating their requests. However, the influence of prompt patterns on code quality has yet to be thoroughly investigated. An improved understanding of this relationship would be essential to advancing our collective knowledge on how to effectively use LLMs for code generation, thereby enhancing their understandability in contemporary software development. This paper empirically investigates the impact of prompt patterns on code quality, specifically maintainability, security, and reliability, using the Dev-GPT dataset. Results show that Zero-Shot prompting is most common, followed by Zero-Shot with Chain-of-Thought and Few-Shot. Analysis of 7583 code files across quality metrics revealed minimal issues, with Kruskal-Wallis tests indicating no significant differences among patterns, suggesting that prompt structure may not substantially impact these quality metrics in ChatGPT-assisted code generation.
Large language models (LLMs) have demonstrated strong capabilities in code generation, underscoring the critical need for rigorous and comprehensive evaluation. Existing evaluation approaches fall into three categories, including human-centered, metric-based, and LLM-based. Considering that human-centered approaches are labour-intensive and metric-based ones overly rely on reference answers, LLM-based approaches are gaining increasing attention due to their stronger contextual understanding capabilities and superior efficiency. However, the performance of LLM-based approaches remains limited due to: (1) lack of multisource domain knowledge, and (2) insufficient comprehension of complex code. To mitigate the limitations, we propose CodeVisionary, the first LLM-based agent framework for evaluating LLMs in code generation. CodeVisionary consists of two stages: (1) Multiscore knowledge analysis stage, which aims to gather multisource and comprehensive domain knowledge by formulating and executing a stepwise evaluation plan. (2) Negotiation-based scoring stage, which involves multiple judges engaging in discussions to better comprehend the complex code and reach a consensus on the evaluation score. Extensive experiments demonstrate that CodeVisionary achieves the best performance for evaluating LLMs in code generation, outperforming the best baseline methods with average improvements of 0.202, 0.139, and 0.117 in Pearson, Spearman, and Kendall-Tau coefficients, respectively. Besides, CodeVisionary provides detailed evaluation reports, which assist developers in identifying shortcomings and making improvements. The resources of CodeVisionary are available at https://anonymous.4open.science/r/CodeVisionary.
A wide range of LM applications require generating text that conforms to syntactic or semantic constraints. Imposing such constraints can be naturally framed as probabilistic conditioning, but exact generation from the resulting distribution -- which can differ substantially from the LM's base distribution -- is generally intractable. In this work, we develop an architecture for controlled LM generation based on sequential Monte Carlo (SMC). Our SMC framework allows us to flexibly incorporate domain- and problem-specific constraints at inference time, and efficiently reallocate computational resources in light of new information during the course of generation. By comparing to a number of alternatives and ablations on four challenging domains -- Python code generation for data science, text-to-SQL, goal inference, and molecule synthesis -- we demonstrate that, with little overhead, our approach allows small open-source language models to outperform models over 8x larger, as well as closed-source, fine-tuned ones. In support of the probabilistic perspective, we show that these performance improvements are driven by better approximation to the posterior distribution. Our system builds on the framework of Lew et al. (2023) and integrates with its language model probabilistic programming language, giving users a simple, programmable way to apply SMC to a broad variety of controlled generation problems.
Chinese-Vicuna is an open-source, resource-efficient language model designed to bridge the gap in Chinese instruction-following capabilities by fine-tuning Meta's LLaMA architecture using Low-Rank Adaptation (LoRA). Targeting low-resource environments, it enables cost-effective deployment on consumer GPUs (e.g., RTX-2080Ti for 7B models) and supports domain-specific adaptation in fields like healthcare and law. By integrating hybrid datasets (BELLE and Guanaco) and 4-bit quantization (QLoRA), the model achieves competitive performance in tasks such as translation, code generation, and domain-specific Q\&A. The project provides a comprehensive toolkit for model conversion, CPU inference, and multi-turn dialogue interfaces, emphasizing accessibility for researchers and developers. Evaluations indicate competitive performance across medical tasks, multi-turn dialogue coherence, and real-time legal updates. Chinese-Vicuna's modular design, open-source ecosystem, and community-driven enhancements position it as a versatile foundation for Chinese LLM applications.
In the rapidly advancing field of robotics, dual-arm coordination and complex object manipulation are essential capabilities for developing advanced autonomous systems. However, the scarcity of diverse, high-quality demonstration data and real-world-aligned evaluation benchmarks severely limits such development. To address this, we introduce RoboTwin, a generative digital twin framework that uses 3D generative foundation models and large language models to produce diverse expert datasets and provide a real-world-aligned evaluation platform for dual-arm robotic tasks. Specifically, RoboTwin creates varied digital twins of objects from single 2D images, generating realistic and interactive scenarios. It also introduces a spatial relation-aware code generation framework that combines object annotations with large language models to break down tasks, determine spatial constraints, and generate precise robotic movement code. Our framework offers a comprehensive benchmark with both simulated and real-world data, enabling standardized evaluation and better alignment between simulated training and real-world performance. We validated our approach using the open-source COBOT Magic Robot platform. Policies pre-trained on RoboTwin-generated data and fine-tuned with limited real-world samples demonstrate significant potential for enhancing dual-arm robotic manipulation systems by improving success rates by over 70% for single-arm tasks and over 40% for dual-arm tasks compared to models trained solely on real-world data.
Despite recent advances in Large Language Models (LLMs) for code generation, the quality of LLM-generated code still faces significant challenges. One significant issue is code repetition, which refers to the model's tendency to generate structurally redundant code, resulting in inefficiencies and reduced readability. To address this, we conduct the first empirical study to investigate the prevalence and nature of repetition across 19 state-of-the-art code LLMs using three widely-used benchmarks. Our study includes both quantitative and qualitative analyses, revealing that repetition is pervasive and manifests at various granularities and extents, including character, statement, and block levels. We further summarize a taxonomy of 20 repetition patterns. Building on our findings, we propose DeRep, a rule-based technique designed to detect and mitigate repetition in generated code. We evaluate DeRep using both open-source benchmarks and in an industrial setting. Our results demonstrate that DeRep significantly outperforms baselines in reducing repetition (with an average improvements of 91.3%, 93.5%, and 79.9% in rep-3, rep-line, and sim-line metrics) and enhancing code quality (with a Pass@1 increase of 208.3% over greedy search). Furthermore, integrating DeRep improves the performance of existing repetition mitigation methods, with Pass@1 improvements ranging from 53.7% to 215.7%.
In this work, we present a benchmark that consists of Jupyter notebooks development trajectories and allows measuring how large language models (LLMs) can leverage runtime information for predicting code output and code generation. We demonstrate that the current generation of LLMs performs poorly on these tasks and argue that there exists a significantly understudied domain in the development of code-based models, which involves incorporating the runtime context.
Large language models (LLMs) have demonstrated significant potential in code generation tasks. However, there remains a performance gap between open-source and closed-source models. To address this gap, existing approaches typically generate large amounts of synthetic data for fine-tuning, which often leads to inefficient training. In this work, we propose a data selection strategy in order to improve the effectiveness and efficiency of training for code-based LLMs. By prioritizing data complexity and ensuring that the sampled subset aligns with the distribution of the original dataset, our sampling strategy effectively selects high-quality data. Additionally, we optimize the tokenization process through a "dynamic pack" technique, which minimizes padding tokens and reduces computational resource consumption. Experimental results show that when training on 40% of the OSS-Instruct dataset, the DeepSeek-Coder-Base-6.7B model achieves an average performance of 66.9%, surpassing the 66.1% performance with the full dataset. Moreover, training time is reduced from 47 minutes to 34 minutes, and the peak GPU memory decreases from 61.47 GB to 42.72 GB during a single epoch. Similar improvements are observed with the CodeLlama-Python-7B model on the Evol-Instruct dataset. By optimizing both data selection and tokenization, our approach not only improves model performance but also improves training efficiency.