Efficient GPU kernels are typically low-level and hardware-specific, and LLM agents can now generate them faster than experts can verify. Existing approaches rely on testing, which provides evidence but not a guarantee of correctness. We introduce LARK, an agentic framework that automatically and formally verifies GPU kernels against their PyTorch reference implementations. LARK proves equivalence in functional and memory behavior under exact arithmetic and establishes a new trust boundary: human effort is only needed to verify operator-level Lean semantics ahead of kernel generation time, instead of checking every single line of generated code. LARK formalizes the GPU machine model and the semantics of each operation in Lean 4 once per target language, and the Lean 4 semantics of any incoming kernel is derived from these pre-built definitions. LLM agents then automate the proof in two steps. First, an agent decomposes the whole kernel semantics into smaller segments and Lean 4 verifies this segmentation, which addresses the intractability of proving equivalence for large kernels directly. Second, agents write Lean proofs that each kernel segment matches its high-level PyTorch counterpart. Together, the two steps give whole-program equivalence. We demonstrate that LARK can be used 1) to ensure correctness in existing code-generation agents and 2) to automatically verify human-written kernel libraries. On the KernelCraft GPU tasks, combined with a simple ReAct agent, LARK enables GLM 5.3 to fully verify 10 of 12 Level-1 kernels and Opus 5.5 all 5 Level-2 kernels. Moreover, over 10 production kernels randomly sampled from popular open-source LLM serving frameworks, LARK verifies 9 and identifies a new data race bug in the tenth. We believe LARK represents a significant step toward verifiable code generation in the agentic era.
Reinforcement learning (RL) has become a dominant paradigm for training large language models (LLMs), particularly for reasoning tasks. Effective RL for LLMs requires massive parallelization and poses an urgent need for efficient training systems. Most existing large-scale RL systems for LLMs are synchronous, alternating generation and training in a batch setting where rollouts in each training batch are generated by the same model. This approach stabilizes RL training but suffers from severe system-level inefficiency: generation must wait until the longest output in the batch is completed before model updates, resulting in GPU underutilization. We present AReaL, a fully asynchronous RL system that completely decouples generation from training. Rollout workers in AReaL continuously generate new outputs without waiting, while training workers update the model whenever a batch of data is collected. AReaL also incorporates a collection of system-level optimizations, leading to substantially higher GPU utilization. To stabilize RL training, AReaL balances the workload of rollout and training workers to control data staleness, and adopts a staleness-enhanced PPO variant to better handle outdated training samples. Extensive experiments on math and code reasoning benchmarks show that AReaL achieves up to 2.77$\times$ training speedup compared to synchronous systems with the same number of GPUs and matched or improved final performance. The code of AReaL is available at https://github.com/inclusionAI/AReaL/.
Designing structured visuals such as presentation slides is essential for communicative needs, necessitating both content creation and visual planning skills. In this work, we tackle the challenge of automated slide generation, where models produce slide presentations from natural language (NL) instructions. We first introduce the SlidesBench benchmark, the first benchmark for slide generation with 7k training and 585 testing examples derived from 310 slide decks across 10 domains. SlidesBench supports evaluations that are (i)reference-based to measure similarity to a target slide, and (ii)reference-free to measure the design quality of generated slides alone. We benchmark end-to-end image generation and program generation methods with a variety of models, and find that programmatic methods produce higher-quality slides in user-interactable formats. Built on the success of program generation, we create AutoPresent, an 8B Llama-based model trained on 7k pairs of instructions paired with code for slide generation, and achieve results comparable to the closed-source model GPT-4o. We further explore iterative design refinement where the model is tasked to self-refine its own output, and we found that this process improves the slide's quality. We hope that our work will provide a basis for future work on generating structured visuals.
Across heterogeneous real-world domains, time-series generation supports simulation, model development, and data augmentation by generating sequences that preserve target-domain statistics and temporal dependencies when real observations are scarce or costly. However, existing methods predominantly address in-domain generation, limiting their reuse in unseen domains. We study reference-conditioned zero-shot time series generation: target domains are unseen during pretraining, generation uses a single target-domain reference sequence, and no model parameters are updated. MarkovTS characterizes sequence evolution through local morphologies and their ordered transitions at multiple temporal scales. Its pretrained encoder quantizes normalized time-series patches into multiscale morphological states with a state codebook, while a transition codebook encodes contextual transitions between adjacent intervals. These transferable state-transition representations condition a pretrained flow-matching model. At generation, the frozen model combines a reference-derived condition with independent Gaussian noise to produce stochastic samples that retain the reference's temporal organization. Across 12 unseen target datasets, MarkovTS outperforms strong baselines in most distributional comparisons and generates diverse, reference-consistent sequences. As downstream augmentation, its samples yield overall gains across forecasting, classification, and anomaly detection benchmarks and diverse model families. Our code is available at https://github.com/blisky-li/MarkovTS.
In addressing the pivotal role of translating natural language queries into SQL commands, we propose a suite of compact, fine-tuned models and self-refine mechanisms to democratize data access and analysis for non-expert users, mitigating risks associated with closed-source Large Language Models. Specifically, we constructed a dataset of over 20K sample for Text-to-SQL as well as the preference dateset, to improve the efficiency in the domain of SQL generation. To further ensure code validity, a code corrector was integrated into the model. Our system, DataGpt-sql, achieved 87.2\% accuracy on the spider-dev, respectively, showcasing the effectiveness of our solution in text-to-SQL conversion tasks. Our code, data, and models are available at \url{https://github.com/CainiaoTechAi/datagpt-sql-7b}
This letter introduces a framework for the automatic generation of hardware cores for Artificial Neural Network (ANN)-based chaotic oscillators. The framework trains the model to approximate a chaotic system, then performs design space exploration yielding potential hardware architectures for its implementation. The framework then generates the corresponding synthesizable High-Level Synthesis code and a validation testbench from a selected solution. The hardware design primarily targets FPGAs. The proposed framework offers a rapid hardware design process of candidate architectures superior to manually designed works in terms of hardware cost and throughput. The source code is available on GitHub.
Large Language Model (LLM) based coding tools have been tremendously successful as software development assistants, yet they are often designed for general purpose programming tasks and perform poorly for more specialized domains such as high performance computing. Creating specialized models and tools for these domains is crucial towards gaining the benefits of LLMs in areas such as HPC. While previous work has explored HPC-specific models, LLMs still struggle to generate parallel code and it is not at all clear what hurdles are still holding back these LLMs and what must be done to overcome them. In this work, we conduct an in-depth study along the many axes of fine-tuning a specialized HPC LLM in order to better understand the challenges. Based on our findings we fine-tune and evaluate a specialized HPC LLM that is shown to be the best performing open-source code LLM for parallel code generation to date.
We present a StyleGAN2-based deep learning approach for 3D shape generation, called SDF-StyleGAN, with the aim of reducing visual and geometric dissimilarity between generated shapes and a shape collection. We extend StyleGAN2 to 3D generation and utilize the implicit signed distance function (SDF) as the 3D shape representation, and introduce two novel global and local shape discriminators that distinguish real and fake SDF values and gradients to significantly improve shape geometry and visual quality. We further complement the evaluation metrics of 3D generative models with the shading-image-based Fr\'echet inception distance (FID) scores to better assess visual quality and shape distribution of the generated shapes. Experiments on shape generation demonstrate the superior performance of SDF-StyleGAN over the state-of-the-art. We further demonstrate the efficacy of SDF-StyleGAN in various tasks based on GAN inversion, including shape reconstruction, shape completion from partial point clouds, single-view image-based shape generation, and shape style editing. Extensive ablation studies justify the efficacy of our framework design. Our code and trained models are available at https://github.com/Zhengxinyang/SDF-StyleGAN.
Commit messages are natural language descriptions of code changes, which are important for software evolution such as code understanding and maintenance. However, previous methods are trained on the entire dataset without considering the fact that a portion of commit messages adhere to good practice (i.e., good-practice commits), while the rest do not. On the basis of our empirical study, we discover that training on good-practice commits significantly contributes to the commit message generation. Motivated by this finding, we propose a novel knowledge-aware denoising learning method called KADEL. Considering that good-practice commits constitute only a small proportion of the dataset, we align the remaining training samples with these good-practice commits. To achieve this, we propose a model that learns the commit knowledge by training on good-practice commits. This knowledge model enables supplementing more information for training samples that do not conform to good practice. However, since the supplementary information may contain noise or prediction errors, we propose a dynamic denoising training method. This method composes a distribution-aware confidence function and a dynamic distribution list, which enhances the effectiveness of the training process. Experimental results on the whole MCMD dataset demonstrate that our method overall achieves state-of-the-art performance compared with previous methods. Our source code and data are available at https://github.com/DeepSoftwareAnalytics/KADEL
Despite extensive usage in high-performance, low-level systems programming applications, C is susceptible to vulnerabilities due to manual memory management and unsafe pointer operations. Rust, a modern systems programming language, offers a compelling alternative. Its unique ownership model and type system ensure memory safety without sacrificing performance. In this paper, we present Syzygy, an automated approach to translate C to safe Rust. Our technique uses a synergistic combination of LLM-driven code and test translation guided by dynamic-analysis-generated execution information. This paired translation runs incrementally in a loop over the program in dependency order of the code elements while maintaining per-step correctness. Our approach exposes novel insights on combining the strengths of LLMs and dynamic analysis in the context of scaling and combining code generation with testing. We apply our approach to successfully translate Zopfli, a high-performance compression library with ~3000 lines of code and 98 functions. We validate the translation by testing equivalence with the source C program on a set of inputs. To our knowledge, this is the largest automated and test-validated C to safe Rust code translation achieved so far.
Large Language Models (LLMs) are increasingly used in post-development tasks such as code repair and testing. A key factor in these tasks' success is the model's deep understanding of code. However, the extent to which LLMs truly understand code remains largely unevaluated. Quantifying code comprehension is challenging due to its abstract nature and the lack of a standardized metric. Previously, this was assessed through developer surveys, which are not feasible for evaluating LLMs. Existing LLM benchmarks focus primarily on code generation, fundamentally different from code comprehension. Additionally, fixed benchmarks quickly become obsolete as they become part of the training data. This paper presents the first large-scale empirical investigation into LLMs' ability to understand code. Inspired by mutation testing, we use an LLM's fault-finding ability as a proxy for its deep code understanding. This approach is based on the insight that a model capable of identifying subtle functional discrepancies must understand the code well. We inject faults in real-world programs and ask the LLM to localize them, ensuring the specifications suffice for fault localization. Next, we apply semantic-preserving code mutations (SPMs) to the faulty programs and test whether the LLMs still locate the faults, verifying their confidence in code understanding. We evaluate nine popular LLMs on 600,010 debugging tasks from 670 Java and 637 Python programs. We find that LLMs lose the ability to debug the same bug in 78% of faulty programs when SPMs are applied, indicating a shallow understanding of code and reliance on features irrelevant to semantics. We also find that LLMs understand code earlier in the program better than later. This suggests that LLMs' code comprehension remains tied to lexical and syntactic features due to tokenization designed for natural languages, which overlooks code semantics.
In this work, we present TOAD-GAN (Token-based One-shot Arbitrary Dimension Generative Adversarial Network), a novel Procedural Content Generation (PCG) algorithm that generates token-based video game levels. TOAD-GAN follows the SinGAN architecture and can be trained using only one example. We demonstrate its application for Super Mario Bros. levels and are able to generate new levels of similar style in arbitrary sizes. We achieve state-of-the-art results in modeling the patterns of the training level and provide a comparison with different baselines under several metrics. Additionally, we present an extension of the method that allows the user to control the generation process of certain token structures to ensure a coherent global level layout. We provide this tool to the community to spur further research by publishing our source code.
As the text generation capabilities of large language models become increasingly prominent, recent studies have focused on controlling particular aspects of the generated text to make it more personalized. However, most research on controllable text generation focuses on controlling the content or modeling specific high-level/coarse-grained attributes that reflect authors' writing styles, such as formality, domain, or sentiment. In this paper, we focus on controlling fine-grained attributes spanning multiple linguistic dimensions, such as lexical and syntactic attributes. We introduce a novel benchmark to train generative models and evaluate their ability to generate personalized text based on multiple fine-grained linguistic attributes. We systematically investigate the performance of various large language models on our benchmark and draw insights from the factors that impact their performance. We make our code, data, and pretrained models publicly available.
This paper presents improved native unified multimodal models, \emph{i.e.,} Show-o2, that leverage autoregressive modeling and flow matching. Built upon a 3D causal variational autoencoder space, unified visual representations are constructed through a dual-path of spatial (-temporal) fusion, enabling scalability across image and video modalities while ensuring effective multimodal understanding and generation. Based on a language model, autoregressive modeling and flow matching are natively applied to the language head and flow head, respectively, to facilitate text token prediction and image/video generation. A two-stage training recipe is designed to effectively learn and scale to larger models. The resulting Show-o2 models demonstrate versatility in handling a wide range of multimodal understanding and generation tasks across diverse modalities, including text, images, and videos. Code and models are released at https://github.com/showlab/Show-o.
Language generation based on maximum likelihood estimation (MLE) has become the fundamental approach for text generation. Maximum likelihood estimation is typically performed by minimizing the log-likelihood loss, also known as the logarithmic score in statistical decision theory. The logarithmic score is strictly proper in the sense that it encourages honest forecasts, where the expected score is maximized only when the model reports true probabilities. Although many strictly proper scoring rules exist, the logarithmic score is the only local scoring rule among them that depends exclusively on the probability of the observed sample, making it capable of handling the exponentially large sample space of natural text. In this work, we propose a straightforward strategy for adapting scoring rules to language generation, allowing for language modeling with any non-local scoring rules. Leveraging this strategy, we train language generation models using two classic strictly proper scoring rules, the Brier score and the Spherical score, as alternatives to the logarithmic score. Experimental results indicate that simply substituting the loss function, without adjusting other hyperparameters, can yield substantial improvements in model's generation capabilities. Moreover, these improvements can scale up to large language models (LLMs) such as LLaMA-7B and LLaMA-13B. Source code: \url{https://github.com/shaochenze/ScoringRulesLM}.
Recent language models have demonstrated proficiency in summarizing source code. However, as in many other domains of machine learning, language models of code lack sufficient explainability. Informally, we lack a formulaic or intuitive understanding of what and how models learn from code. Explainability of language models can be partially provided if, as the models learn to produce higher-quality code summaries, they also align in deeming the same code parts important as those identified by human programmers. In this paper, we report negative results from our investigation of explainability of language models in code summarization through the lens of human comprehension. We measure human focus on code using eye-tracking metrics such as fixation counts and duration in code summarization tasks. To approximate language model focus, we employ a state-of-the-art model-agnostic, black-box, perturbation-based approach, SHAP (SHapley Additive exPlanations), to identify which code tokens influence that generation of summaries. Using these settings, we find no statistically significant relationship between language models' focus and human programmers' attention. Furthermore, alignment between model and human foci in this setting does not seem to dictate the quality of the LLM-generated summaries. Our study highlights an inability to align human focus with SHAP-based model focus measures. This result calls for future investigation of multiple open questions for explainable language models for code summarization and software engineering tasks in general, including the training mechanisms of language models for code, whether there is an alignment between human and model attention on code, whether human attention can improve the development of language models, and what other model focus measures are appropriate for improving explainability.
Image generation has witnessed significant advancements in the past few years. However, evaluating the performance of image generation models remains a formidable challenge. In this paper, we propose ICE-Bench, a unified and comprehensive benchmark designed to rigorously assess image generation models. Its comprehensiveness could be summarized in the following key features: (1) Coarse-to-Fine Tasks: We systematically deconstruct image generation into four task categories: No-ref/Ref Image Creating/Editing, based on the presence or absence of source images and reference images. And further decompose them into 31 fine-grained tasks covering a broad spectrum of image generation requirements, culminating in a comprehensive benchmark. (2) Multi-dimensional Metrics: The evaluation framework assesses image generation capabilities across 6 dimensions: aesthetic quality, imaging quality, prompt following, source consistency, reference consistency, and controllability. 11 metrics are introduced to support the multi-dimensional evaluation. Notably, we introduce VLLM-QA, an innovative metric designed to assess the success of image editing by leveraging large models. (3) Hybrid Data: The data comes from real scenes and virtual generation, which effectively improves data diversity and alleviates the bias problem in model evaluation. Through ICE-Bench, we conduct a thorough analysis of existing generation models, revealing both the challenging nature of our benchmark and the gap between current model capabilities and real-world generation requirements. To foster further advancements in the field, we will open-source ICE-Bench, including its dataset, evaluation code, and models, thereby providing a valuable resource for the research community.
Automated generation of clinically accurate radiology reports can improve patient care. Previous report generation methods that rely on image captioning models often generate incoherent and incorrect text due to their lack of relevant domain knowledge, while retrieval-based attempts frequently retrieve reports that are irrelevant to the input image. In this work, we propose Contrastive X-Ray REport Match (X-REM), a novel retrieval-based radiology report generation module that uses an image-text matching score to measure the similarity of a chest X-ray image and radiology report for report retrieval. We observe that computing the image-text matching score with a language-image model can effectively capture the fine-grained interaction between image and text that is often lost when using cosine similarity. X-REM outperforms multiple prior radiology report generation modules in terms of both natural language and clinical metrics. Human evaluation of the generated reports suggests that X-REM increased the number of zero-error reports and decreased the average error severity compared to the baseline retrieval approach. Our code is available at: https://github.com/rajpurkarlab/X-REM
In the field of sketch generation, raster-format trained models often produce non-stroke artifacts, while vector-format trained models typically lack a holistic understanding of sketches, leading to compromised recognizability. Moreover, existing methods struggle to extract common features from similar elements (e.g., eyes of animals) appearing at varying positions across sketches. To address these challenges, we propose StrokeFusion, a two-stage framework for vector sketch generation. It contains a dual-modal sketch feature learning network that maps strokes into a high-quality latent space. This network decomposes sketches into normalized strokes and jointly encodes stroke sequences with Unsigned Distance Function (UDF) maps, representing sketches as sets of stroke feature vectors. Building upon this representation, our framework exploits a stroke-level latent diffusion model that simultaneously adjusts stroke position, scale, and trajectory during generation. This enables high-fidelity sketch generation while supporting stroke interpolation editing. Extensive experiments on the QuickDraw dataset demonstrate that our framework outperforms state-of-the-art techniques, validating its effectiveness in preserving structural integrity and semantic features. Code and models will be made publicly available upon publication.
Text-to-image (T2I) generation has made remarkable progress, yet existing systems still lack intuitive control over spatial composition, object consistency, and multi-step editing. We present $\textbf{LayerCraft}$, a modular framework that uses large language models (LLMs) as autonomous agents to orchestrate structured, layered image generation and editing. LayerCraft supports two key capabilities: (1) $\textit{structured generation}$ from simple prompts via chain-of-thought (CoT) reasoning, enabling it to decompose scenes, reason about object placement, and guide composition in a controllable, interpretable manner; and (2) $\textit{layered object integration}$, allowing users to insert and customize objects -- such as characters or props -- across diverse images or scenes while preserving identity, context, and style. The system comprises a coordinator agent, the $\textbf{ChainArchitect}$ for CoT-driven layout planning, and the $\textbf{Object Integration Network (OIN)}$ for seamless image editing using off-the-shelf T2I models without retraining. Through applications like batch collage editing and narrative scene generation, LayerCraft empowers non-experts to iteratively design, customize, and refine visual content with minimal manual effort. Code will be released at https://github.com/PeterYYZhang/LayerCraft.