PyTorchKR Papers

“Code Generation”

총 8,436편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2503.23899

Rubrik's Cube: Testing a New Rubric for Evaluating Explanations on the CUBE dataset

The performance and usability of Large-Language Models (LLMs) are driving their use in explanation generation tasks. However, despite their widespread adoption, LLM explanations have been found to be unreliable, making it difficult for users to distinguish good from bad explanations. To address this issue, we present Rubrik's CUBE, an education-inspired rubric and a dataset of 26k explanations, written and later quality-annotated using the rubric by both humans and six open- and closed-source LLMs. The CUBE dataset focuses on two reasoning and two language tasks, providing the necessary diversity for us to effectively test our proposed rubric. Using Rubrik, we find that explanations are influenced by both task and perceived difficulty. Low quality stems primarily from a lack of conciseness in LLM-generated explanations, rather than cohesion and word choice. The full dataset, rubric, and code will be made available upon acceptance.

arXiv:2504.12737

Chinese-Vicuna: A Chinese Instruction-following Llama-based Model

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.

arXiv:2505.07834

ai.txt: A Domain-Specific Language for Guiding AI Interactions with the Internet

We introduce ai.txt, a novel domain-specific language (DSL) designed to explicitly regulate interactions between AI models, agents, and web content, addressing critical limitations of the widely adopted robots.txt standard. As AI increasingly engages with online materials for tasks such as training, summarization, and content modification, existing regulatory methods lack the necessary granularity and semantic expressiveness to ensure ethical and legal compliance. ai.txt extends traditional URL-based access controls by enabling precise element-level regulations and incorporating natural language instructions interpretable by AI systems. To facilitate practical deployment, we provide an integrated development environment with code autocompletion and automatic XML generation. Furthermore, we propose two compliance mechanisms: XML-based programmatic enforcement and natural language prompt integration, and demonstrate their effectiveness through preliminary experiments and case studies. Our approach aims to aid the governance of AI-Internet interactions, promoting responsible AI use in digital ecosystems.

arXiv:2505.12200

CompBench: Benchmarking Complex Instruction-guided Image Editing

While real-world applications increasingly demand intricate scene manipulation, existing instruction-guided image editing benchmarks often oversimplify task complexity and lack comprehensive, fine-grained instructions. To bridge this gap, we introduce, a large-scale benchmark specifically designed for complex instruction-guided image editing. CompBench features challenging editing scenarios that incorporate fine-grained instruction following, spatial and contextual reasoning, thereby enabling comprehensive evaluation of image editing models' precise manipulation capabilities. To construct CompBench, We propose an MLLM-human collaborative framework with tailored task pipelines. Furthermore, we propose an instruction decoupling strategy that disentangles editing intents into four key dimensions: location, appearance, dynamics, and objects, ensuring closer alignment between instructions and complex editing requirements. Extensive evaluations reveal that CompBench exposes fundamental limitations of current image editing models and provides critical insights for the development of next-generation instruction-guided image editing systems. The dataset, code, and models are available in https://comp-bench.github.io/.

arXiv:2506.14681

Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality

Supervised fine-tuning (SFT) is a critical step in aligning large language models (LLMs) with human instructions and values, yet many aspects of SFT remain poorly understood. We trained a wide range of base models on a variety of datasets including code generation, mathematical reasoning, and general-domain tasks, resulting in 1,000+ SFT models under controlled conditions. We then identified the dataset properties that matter most and examined the layer-wise modifications introduced by SFT. Our findings reveal that some training-task synergies persist across all models while others vary substantially, emphasizing the importance of model-specific strategies. Moreover, we demonstrate that perplexity consistently predicts SFT effectiveness--often surpassing superficial similarity between trained data and benchmark--and that mid-layer weight changes correlate most strongly with performance gains. We will release these 1,000+ SFT models and benchmark results to accelerate further research.

arXiv:2610.37912v1eess.AS

Beyond Utterance-Level Speaker Similarity: Perception-Aligned Segmental Evaluation of Accent in Speech Generation

Utterance-level similarity ratings do not reveal where perceived accent differences occur in generated speech. We introduce a segmental evaluation protocol combining speaker and accent similarity ratings with subword annotations of perceived accent differences provided by listeners from target accent regions. Using newly collected recordings, 47 retained listeners evaluate 200 content-matched reference--candidate pairs spanning four Scottish and Australian voices, two zero-shot TTS systems, and an articulation-based voice conversion system. Subword annotations locate salient pronunciation differences and sometimes yield different system rankings from similarity ratings. Speaker and accent ratings remain strongly associated ($\rho = 0.71$), highlighting the value of examining local perceptual evidence beyond these scores. Evaluation of seven objective distances reveals weak correspondence with subword selections, with correlations reaching at most $\rho = 0.22$. We release code, data, and an interactive demo to support the development of perceptually grounded segmental accent evaluation.

arXiv:2611.09423v1cs.AI

Agentic AI for Healthcare Claim Denial Analysis: A Policy-Grounded Framework for Privacy Preservation

In the healthcare sector, claim denials impose significant financial and administrative burdens, emphasizing the need for automated and trustworthy decision-support systems. This paper proposes a modular Agentic Artificial Intelligence (AI) framework for healthcare claim denial analysis, integrating specialized agents for denial risk prediction, risk factor identification, diagnosis-procedure-medication code relationship analysis, and counterfactual analysis. The framework integrates predictive analytics, graph-based learning, a Hybrid Contextual Retrieval-Augmented Generation (RAG) pipeline grounded in healthcare policy documents, and a dedicated HIPAA Compliance Agent to support privacy protection. Evaluated on a synthetic healthcare claims dataset, the framework achieved 98% traceability of reported numerical findings, validated 214 of 216 policy citations, and removed all 280 identified sensitive identifiers without modifying the underlying analysis. These findings pave the way for further research into Agentic AI as a framework for explainable, policy-aware, and privacy-preserving healthcare claim denial management.

arXiv:2205.05019공식 코드 · Papers with Code

Learning to Answer Visual Questions from Web Videos

Recent methods for visual question answering rely on large-scale annotated datasets. Manual annotation of questions and answers for videos, however, is tedious, expensive and prevents scalability. In this work, we propose to avoid manual annotation and generate a large-scale training dataset for video question answering making use of automatic cross-modal supervision. We leverage a question generation transformer trained on text data and use it to generate question-answer pairs from transcribed video narrations. Given narrated videos, we then automatically generate the HowToVQA69M dataset with 69M video-question-answer triplets. To handle the open vocabulary of diverse answers in this dataset, we propose a training procedure based on a contrastive loss between a video-question multi-modal transformer and an answer transformer. We introduce the zero-shot VideoQA task and the VideoQA feature probe evaluation setting and show excellent results, in particular for rare answers. Furthermore, our method achieves competitive results on MSRVTT-QA, ActivityNet-QA, MSVD-QA and How2QA datasets. We also show that our VideoQA dataset generation approach generalizes to another source of web video and text data. We use our method to generate the WebVidVQA3M dataset from the WebVid dataset, i.e., videos with alt-text annotations, and show its benefits for training VideoQA models. Finally, for a detailed evaluation we introduce iVQA, a new VideoQA dataset with reduced language bias and high-quality manual annotations. Code, datasets and trained models are available at https://antoyang.github.io/just-ask.html

arXiv:2211.09460공식 코드 · Papers with Code

Progressive Tree-Structured Prototype Network for End-to-End Image Captioning

Studies of image captioning are shifting towards a trend of a fully end-to-end paradigm by leveraging powerful visual pre-trained models and transformer-based generation architecture for more flexible model training and faster inference speed. State-of-the-art approaches simply extract isolated concepts or attributes to assist description generation. However, such approaches do not consider the hierarchical semantic structure in the textual domain, which leads to an unpredictable mapping between visual representations and concept words. To this end, we propose a novel Progressive Tree-Structured prototype Network (dubbed PTSN), which is the first attempt to narrow down the scope of prediction words with appropriate semantics by modeling the hierarchical textual semantics. Specifically, we design a novel embedding method called tree-structured prototype, producing a set of hierarchical representative embeddings which capture the hierarchical semantic structure in textual space. To utilize such tree-structured prototypes into visual cognition, we also propose a progressive aggregation module to exploit semantic relationships within the image and prototypes. By applying our PTSN to the end-to-end captioning framework, extensive experiments conducted on MSCOCO dataset show that our method achieves a new state-of-the-art performance with 144.2% (single model) and 146.5% (ensemble of 4 models) CIDEr scores on `Karpathy' split and 141.4% (c5) and 143.9% (c40) CIDEr scores on the official online test server. Trained models and source code have been released at: https://github.com/NovaMind-Z/PTSN.

arXiv:2305.00942공식 코드 · Papers with Code

StyleAvatar: Real-time Photo-realistic Portrait Avatar from a Single Video

Face reenactment methods attempt to restore and re-animate portrait videos as realistically as possible. Existing methods face a dilemma in quality versus controllability: 2D GAN-based methods achieve higher image quality but suffer in fine-grained control of facial attributes compared with 3D counterparts. In this work, we propose StyleAvatar, a real-time photo-realistic portrait avatar reconstruction method using StyleGAN-based networks, which can generate high-fidelity portrait avatars with faithful expression control. We expand the capabilities of StyleGAN by introducing a compositional representation and a sliding window augmentation method, which enable faster convergence and improve translation generalization. Specifically, we divide the portrait scenes into three parts for adaptive adjustments: facial region, non-facial foreground region, and the background. Besides, our network leverages the best of UNet, StyleGAN and time coding for video learning, which enables high-quality video generation. Furthermore, a sliding window augmentation method together with a pre-training strategy are proposed to improve translation generalization and training performance, respectively. The proposed network can converge within two hours while ensuring high image quality and a forward rendering time of only 20 milliseconds. Furthermore, we propose a real-time live system, which further pushes research into applications. Results and experiments demonstrate the superiority of our method in terms of image quality, full portrait video generation, and real-time re-animation compared to existing facial reenactment methods. Training and inference code for this paper are at https://github.com/LizhenWangT/StyleAvatar.

arXiv:2306.11363공식 코드 · Papers with Code

Masked Diffusion Models Are Fast Distribution Learners

Diffusion model has emerged as the \emph{de-facto} model for image generation, yet the heavy training overhead hinders its broader adoption in the research community. We observe that diffusion models are commonly trained to learn all fine-grained visual information from scratch. This paradigm may cause unnecessary training costs hence requiring in-depth investigation. In this work, we show that it suffices to train a strong diffusion model by first pre-training the model to learn some primer distribution that loosely characterizes the unknown real image distribution. Then the pre-trained model can be fine-tuned for various generation tasks efficiently. In the pre-training stage, we propose to mask a high proportion (e.g., up to 90\%) of input images to approximately represent the primer distribution and introduce a masked denoising score matching objective to train a model to denoise visible areas. In subsequent fine-tuning stage, we efficiently train diffusion model without masking. Utilizing the two-stage training framework, we achieves significant training acceleration and a new FID score record of 6.27 on CelebA-HQ $256 \times 256$ for ViT-based diffusion models. The generalizability of a pre-trained model further helps building models that perform better than ones trained from scratch on different downstream datasets. For instance, a diffusion model pre-trained on VGGFace2 attains a 46\% quality improvement when fine-tuned on a different dataset that contains only 3000 images. Our code is available at \url{https://github.com/jiachenlei/maskdm}.

arXiv:2310.08166

Ziya-Visual: Bilingual Large Vision-Language Model via Multi-Task Instruction Tuning

Recent advancements enlarge the capabilities of large language models (LLMs) in zero-shot image-to-text generation and understanding by integrating multi-modal inputs. However, such success is typically limited to English scenarios due to the lack of large-scale and high-quality non-English multi-modal resources, making it extremely difficult to establish competitive counterparts in other languages. In this paper, we introduce the Ziya-Visual series, a set of bilingual large-scale vision-language models (LVLMs) designed to incorporate visual semantics into LLM for multi-modal dialogue. Composed of Ziya-Visual-Base and Ziya-Visual-Chat, our models adopt the Querying Transformer from BLIP-2, further exploring the assistance of optimization schemes such as instruction tuning, multi-stage training and low-rank adaptation module for visual-language alignment. In addition, we stimulate the understanding ability of GPT-4 in multi-modal scenarios, translating our gathered English image-text datasets into Chinese and generating instruction-response through the in-context learning method. The experiment results demonstrate that compared to the existing LVLMs, Ziya-Visual achieves competitive performance across a wide range of English-only tasks including zero-shot image-text retrieval, image captioning, and visual question answering. The evaluation leaderboard accessed by GPT-4 also indicates that our models possess satisfactory image-text understanding and generation capabilities in Chinese multi-modal scenario dialogues. Code, demo and models are available at ~\url{https://huggingface.co/IDEA-CCNL/Ziya-BLIP2-14B-Visual-v1}.

arXiv:2403.17752공식 코드 · Papers with Code

Can multiple-choice questions really be useful in detecting the abilities of LLMs?

Multiple-choice questions (MCQs) are widely used in the evaluation of large language models (LLMs) due to their simplicity and efficiency. However, there are concerns about whether MCQs can truly measure LLM's capabilities, particularly in knowledge-intensive scenarios where long-form generation (LFG) answers are required. The misalignment between the task and the evaluation method demands a thoughtful analysis of MCQ's efficacy, which we undertake in this paper by evaluating nine LLMs on four question-answering (QA) datasets in two languages: Chinese and English. We identify a significant issue: LLMs exhibit an order sensitivity in bilingual MCQs, favoring answers located at specific positions, i.e., the first position. We further quantify the gap between MCQs and long-form generation questions (LFGQs) by comparing their direct outputs, token logits, and embeddings. Our results reveal a relatively low correlation between answers from MCQs and LFGQs for identical questions. Additionally, we propose two methods to quantify the consistency and confidence of LLMs' output, which can be generalized to other QA evaluation benchmarks. Notably, our analysis challenges the idea that the higher the consistency, the greater the accuracy. We also find MCQs to be less reliable than LFGQs in terms of expected calibration error. Finally, the misalignment between MCQs and LFGQs is not only reflected in the evaluation performance but also in the embedding space. Our code and models can be accessed at https://github.com/Meetyou-AI-Lab/Can-MC-Evaluate-LLMs.

논문 ID: 9233.36393

Exploring Naive Approaches to Tell Apart LLMs Productions from Human-written Text

Powerful Large Language Models (large LMs or LLMs) such as BERT and GPT are making the task of detecting machine-generated text more and more prominent and crucial to minimize threats posed by text generation models misuse. Nonetheless, only a limited number of efforts exist so far, which can be classified into simple classifiers, zero-shot approaches, and fine-tuned LMs. These approaches usually rely on LMs whose discrimination accuracy decreases as the size difference in favor of the generator model increases (hence, a detector should always employ a LM with at least the same number of parameters of the source LM). Also, most of these approaches do not explicitly investigate whether the sentence syntactic structure can provide additional information that helps to build better detectors. All these considerations make the generalizing ability of detection methods into question. While generation techniques become more and more capable of producing human-like text, are the detection techniques capable of keeping up if not properly trained? In this paper, we evaluate the most effective (and reproducible) detection method available in the state of the art in order to figure out the limits in its robustness. We complement this analysis by discussing results obtained using a novel naive approach that demonstrably achieves comparable results in terms of robustness with respect to much more advanced and sophisticated state-of-the-art methods. Code with details on experiments are available at: https://github.com/bancaditalia/gen-text-detect.

arXiv:2608.14011v2cs.IR

EchoRec: Multi-Item Prediction-Empowered Generative Recommendation via Cycle-Consistent Preference Alignment

Generative recommendation autoregressively generates the semantic IDs of the target item, unifying preference modeling and index retrieval within the shared token space. Recent attempts have introduced Multi-Token Prediction (MTP) into this field, yet they primarily inherit its efficiency merit, leaving its potential as dense supervision unexplored. Unlocking this potential hinges on whether future behaviors qualify as informative supervision. Our analysis reveals that future behaviors carry a semantic echo of the current one far above that of random pairs, which nevertheless decays along horizons under intent transitions, making them informative yet order-dependent signals. Motivated by this, we propose EchoRec, which empowers MTP with cycle-consistent holistic preference alignment across multi-horizon for generative recommendation. It comprises two synergistic modules. Horizon-aware Preference Generation (HPG) sequentially chains lightweight auxiliary branches upon the base recommender, where each branch conditions on its predecessor to respect preference evolution. Verifiable Holistic-Preference Alignment (VHA) further consolidates them into the holistic preference and echoes it back through cycle-consistent projectors to suppress spurious alignment, with a theoretical guarantee that excludes the rank-collapse form of spurious alignment under an invertible round trip, enabling the holistic preference to be retained in the decoding representation. All auxiliary components serve as disposable scaffolding discarded at inference, introducing negligible online serving overhead. Extensive experiments on four datasets demonstrate the superiority of our EchoRec, together with its naturally acquired multi-item generation ability. Our code and datasets will be available upon acceptance.

arXiv:2606.06252v2cs.AI

Can Test-Time Reconstruction Close the Loop on Latent Reasoning?

Recent work moves intermediate reasoning from natural-language traces into latent or cache-level representations to reduce token overhead and avoid a discrete communication bottleneck. However, this shift also removes a key advantage of textual reasoning: intermediate states are no longer inspectable, making it difficult to determine whether a latent state still preserves the constraints of the original query. As a result, latent reasoning typically operates in an open loop, where a latent state is produced and consumed without an input-anchored fidelity check. We propose ReLAT (Reconstruction-Guided Latent Reasoning At Test Time), a self-supervised test-time training method that closes this loop using the query itself as the reference. Our key observation is that if a latent state faithfully represents a query, the query should be recoverable from it; if the query cannot be recovered, the latent state has lost task-relevant information. ReLAT operationalizes this principle by constructing a differentiable Question -> Latent Thought -> Question cycle and optimizing query reconstruction loss through the latent thought before answer generation. This anchors opaque latent computation to the problem specification it is supposed to represent. Across mathematical reasoning, knowledge QA, and code generation benchmarks on the Qwen family, ReLAT consistently improves over single-model inference, text-based collaboration, open-loop latent collaboration, and alternative test-time training objectives. On Qwen3-8B, ReLAT raises AIME 2024 accuracy from 56.7% to 73.3%, a 16.6-point gain over the strongest open-loop latent baseline.

arXiv:2511.20104v3cs.LG

The Devil in the Details: Emergent Misalignment, Format and Coherence in Open-Weights LLMs

Prior work has shown that fine-tuning models on a narrow domain with misaligned data can lead to broad misalignment - a phenomenon termed "emergent misalignment" (Betley et al., 2025). While all tested models were susceptible to emergent misalignment, some models showed more resistance than others. Specifically the Qwen-2.5 family proved to be relatively resistant, while GPT-4o exhibited the strongest misalignment. In this paper we evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family and measure misalignment robustness over a range of model architectures and scales. We replicate the effect across nine modern open-weights models (Gemma 3 and Qwen 3 families, 1B-32B parameters). Models fine-tuned on insecure code generation show a 0.74% misalignment rate (compared to 0.07% for base models), matching the lower end of prior open-model results but dramatically lower than GPT-4o's 20%. We identify a critical format-dependent vulnerability: requiring JSON output doubles misalignment rates compared to natural language prompts (0.96% vs 0.42%), an effect concentrated in the Gemma 3 family (1.43% vs 0.22%) and not significant in Qwen 3 (0.72% vs 0.59%). This suggests that structural constraints may bypass safety training by reducing the model's 'degrees of freedom' to refuse. These findings confirm emergent misalignment as a reproducible phenomenon in modern open-weights models, with rates substantially lower than observed in proprietary systems.

arXiv:2407.02813공식 코드 · Papers with Code

Data Overfitting for On-Device Super-Resolution with Dynamic Algorithm and Compiler Co-Design

Deep neural networks (DNNs) are frequently employed in a variety of computer vision applications. Nowadays, an emerging trend in the current video distribution system is to take advantage of DNN's overfitting properties to perform video resolution upscaling. By splitting videos into chunks and applying a super-resolution (SR) model to overfit each chunk, this scheme of SR models plus video chunks is able to replace traditional video transmission to enhance video quality and transmission efficiency. However, many models and chunks are needed to guarantee high performance, which leads to tremendous overhead on model switching and memory footprints at the user end. To resolve such problems, we propose a Dynamic Deep neural network assisted by a Content-Aware data processing pipeline to reduce the model number down to one (Dy-DCA), which helps promote performance while conserving computational resources. Additionally, to achieve real acceleration on the user end, we designed a framework that optimizes dynamic features (e.g., dynamic shapes, sizes, and control flow) in Dy-DCA to enable a series of compilation optimizations, including fused code generation, static execution planning, etc. By employing such techniques, our method achieves better PSNR and real-time performance (33 FPS) on an off-the-shelf mobile phone. Meanwhile, assisted by our compilation optimization, we achieve a 1.7$\times$ speedup while saving up to 1.61$\times$ memory consumption. Code available in https://github.com/coulsonlee/Dy-DCA-ECCV2024.

arXiv:2001.10161공식 코드 · Papers with Code

Bringing Stories Alive: Generating Interactive Fiction Worlds

World building forms the foundation of any task that requires narrative intelligence. In this work, we focus on procedurally generating interactive fiction worlds---text-based worlds that players "see" and "talk to" using natural language. Generating these worlds requires referencing everyday and thematic commonsense priors in addition to being semantically consistent, interesting, and coherent throughout. Using existing story plots as inspiration, we present a method that first extracts a partial knowledge graph encoding basic information regarding world structure such as locations and objects. This knowledge graph is then automatically completed utilizing thematic knowledge and used to guide a neural language generation model that fleshes out the rest of the world. We perform human participant-based evaluations, testing our neural model's ability to extract and fill-in a knowledge graph and to generate language conditioned on it against rule-based and human-made baselines. Our code is available at https://github.com/rajammanabrolu/WorldGeneration.

arXiv:2107.08792

Selective Focusing Learning in Conditional GANs

Conditional generative adversarial networks (cGANs) have demonstrated remarkable success due to their class-wise controllability and superior quality for complex generation tasks. Typical cGANs solve the joint distribution matching problem by decomposing two easier sub-problems: marginal matching and conditional matching. From our toy experiments, we found that it is the best to apply only conditional matching to certain samples due to the content-aware optimization of the discriminator. This paper proposes a simple (a few lines of code) but effective training methodology, selective focusing learning, which enforces the discriminator and generator to learn easy samples of each class rapidly while maintaining diversity. Our key idea is to selectively apply conditional and joint matching for the data in each mini-batch. We conducted experiments on recent cGAN variants in ImageNet (64x64 and 128x128), CIFAR-10, and CIFAR-100 datasets, and improved the performance significantly (up to 35.18% in terms of FID) without sacrificing diversity.