🔥 Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation

💡 The paper introduces Explorative Modeling, a new paradigm that unlocks a third pretraining axis for existing generative models, beyond parameters and data. The key problem addressed is that generative models are not trained end-to-end, due to the fact that they handle distributions with many modes, and existing scalable approaches factor the generation procedure, preventing end-to-end generation.

The proposed method, Explorative Modeling, factors the training loop instead, exploring K candidate matches between model generations and data, and training on the best predictions, which commit to modes rather than blurring them. This approach is found to be useful in two settings.

First, increasing exploration adds a third pretraining axis for existing generative models, where scaling exploration monotonically improves performance across both continuous and discrete domains, such as images, video, and language. Notably, gains from exploration increase with scale, climbing from 7 percent to 36 percent as data scales and from 13 percent to 23 percent as models grow, with efficiency gains more than doubling at three times the compute.

Second, Explorative Models enable end-to-end reconstructive generative modeling, matching diffusion on control tasks with 16-256 times fewer inference steps. The results establish Explorative Models as both a new pretraining axis for existing generative models and a standalone end-to-end generative modeling paradigm.

Overall, the paper's contributions include improving FLOP efficiency by 4.1 times, sample efficiency by 6.2 times, and parameter efficiency by 47 percent, and lifting the strongest image-generation recipes to near state-of-the-art 1.43 FID on ImageNet without guidance, enabling scaling and generalization.

📅 Published on Jul 29

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.27372
• PDF: https://arxiv.org/pdf/2607.27372
• Project Page: https://explorative-modeling.github.io/

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#ExplorativeModeling #GenerativeModels #EndToEndGeneration #PretrainingTechniques #ModeDiscovery

The first bot in the world of Telegram that offers free courses, free certificates. https://xn--r1a.website/UdemySybot

🔥 Amphion: An Open-Source Audio, Music and Speech Generation Toolkit

💡 The paper introduces Amphion, an open source toolkit for audio, music, and speech generation. The goal of Amphion is to support reproducible research and help junior researchers and engineers get started in the field. The toolkit provides model visualizations, which are beneficial for understanding classic models and architectures. Amphion also includes vocoders, which are essential for producing high quality audio signals, and evaluation metrics, which ensure consistent metrics in generation tasks. The toolkit is designed to support individual generation tasks and offers a platform for studying the conversion of any inputs into general audio. The key contribution of Amphion is its ability to provide a comprehensive platform for audio, music, and speech generation, making it easier for researchers to conduct reproducible research and for junior researchers to enter the field. The paper provides a high level overview of Amphion, highlighting its features and capabilities, and demonstrating its potential to support research and development in audio, music, and speech generation.

📅 Published on Dec 15, 2023

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2312.09911
• PDF: https://arxiv.org/pdf/2312.09911

📊 Datasets citing this paper:
• https://huggingface.co/datasets/Pendrokar/open_tts_tracker

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#AudioGenerationTools #MusicGenerationTechniques #SpeechSynthesisSystems #OpenSourceAudioToolkits #VocoderTechnology

🔥 Transcription Policy as a Latent Variable: Activating Controllable Verbatim ASR with Word-Level Timing

💡 The paper addresses the issue of transcription policy as a latent variable in automatic speech recognition models, which can cause decoding instability, evaluation confounding, and unreliable word-level timing. Modern speech recognition models are trained on heterogeneously annotated data, treating transcription style as an uncontrolled latent variable. The authors show that models already encode both verbatim and intended transcription styles, but the challenge lies in controlled activation.

To address this issue, the authors propose using a coverage-aware decoder task with tokens trained on parallel verbatim and intended transcript pairs. This approach raises German disfluency F1 from 10 to 79 percent zero-shot, despite English-only training. Full English-only fine-tuning surpasses all baselines in verbatim accuracy, disfluency detection, and intended-mode quality across both languages.

The authors further introduce supervised cross-attention fine-tuning, which improves word-level timestamps on disfluent speech beyond forced-alignment baselines. Finally, they propose Verbatimize, a new task enabling scalable creation and enrichment of speech corpora with high-quality canonical verbatim transcripts. The results demonstrate significant improvements in speech recognition and transcription quality, with potential applications in speech processing and related fields.

📅 Published on Jul 21

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.18934
• PDF: https://arxiv.org/pdf/2607.18934
• Project Page: https://nyra-labs.com

🤖 Models citing this paper:
• https://huggingface.co/nyralabs/CrisperWhisper2.0_large
• https://huggingface.co/nyralabs/CrisperWhisper2.0_small
• https://huggingface.co/nyralabs/CrisperWhisper2.0_turbo

🚀 Spaces citing this paper:
• https://huggingface.co/spaces/nyralabs/crisperwhisper2-controllable-asr

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#AutomaticSpeechRecognition #TranscriptionPolicy #LatentVariablesInASR #WordLevelTiming #ControllableASR

🔥 GraphGen: Enhancing Supervised Fine-Tuning for LLMs with Knowledge-Driven Synthetic Data Generation

💡 The paper introduces GraphGen, a framework designed to improve supervised fine-tuning for large language models by generating high-quality synthetic data. The problem addressed is that fine-tuning large language models requires substantial amounts of high-quality supervised data, which is costly and labor-intensive to acquire. Existing synthetic data generation approaches often suffer from factual inaccuracies, insufficient long-tail coverage, simplistic knowledge structures, and homogenized outputs.

To address these challenges, GraphGen constructs a fine-grained knowledge graph from the source text and identifies knowledge gaps in large language models using the expected calibration error metric. It prioritizes the generation of question-answering pairs that target high-value, long-tail knowledge. GraphGen also incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting question-answering data.

The framework is designed for three key question-answering scenarios: atomic question-answering, aggregated question-answering, and multi-hop question-answering. Experimental results on knowledge-intensive tasks under closed-book settings demonstrate that GraphGen outperforms conventional synthetic data methods, offering a more reliable and comprehensive solution to the data scarcity challenge in supervised fine-tuning. The code and data are publicly available, making it a valuable resource for the research community. Overall, GraphGen provides a novel approach to synthetic data generation, addressing the limitations of existing methods and improving the performance of large language models.

📅 Published on May 26, 2025

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2505.20416
• PDF: https://arxiv.org/pdf/2505.20416
• Project Page: https://huggingface.co/spaces/chenzihong/GraphGen

📊 Datasets citing this paper:
• https://huggingface.co/datasets/chenzihong/GraphGen-Data

🚀 Spaces citing this paper:
• https://huggingface.co/spaces/chenzihong/GraphGen

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#KnowledgeGraphConstruction #SyntheticDataGeneration #LargeLanguageModels #SupervisedFineTuning #LLMSupervision

🔥 Scaling Properties of Text Conditioning in Visual Generation

💡 This paper studies the scaling properties of text conditioning in visual generation, which has rarely been measured due to the difficulty of scaling diffusion loss with the number of tokens in natural language prompts. The authors surprisingly find that the converged diffusion loss scales with the amount of structured language in the prompt. To quantify structured language, they adapt two complementary measures: a white-box like likelihood metric and a black-box attribute metric.

The authors use these metrics to analyze the relationship between the converged diffusion loss and the amount of structured language in the prompt. They find that the converged diffusion loss decreases approximately linearly with the white-box metric and follows a power law with the black-box metric across controlled training runs.

Guided by these scaling properties, the authors improve the diffusability of visual generation models by constructing structured prompts with semantic and geometric annotations derived from images. They also improve promptability by training a prompter through supervised fine-tuning, cold-start, and verifier-gated on-policy distillation.

The resulting system outperforms all evaluated open-weight models on nearly every compositional, reasoning, and world knowledge benchmark, while matching or surpassing the strongest closed-weight models on most evaluations. The paper's contributions include providing a better understanding of the scaling properties of text conditioning in visual generation and developing methods to improve the diffusability and promptability of visual generation models.

📅 Published on Jul 31

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.29679
• PDF: https://arxiv.org/pdf/2607.29679
• Project Page: https://heheyas.github.io/context-scaling/

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#VisualGeneration #TextConditioning #DiffusionLoss #NaturalLanguageProcessing #StructuredLanguageMetrics

🔥 Decompile-Bench: Million-Scale Binary-Source Function Pairs for Real-World Binary Decompilation

💡 The paper introduces Decompile-Bench, a large-scale open-source dataset designed to improve the accuracy of large language model-based decompilers. The problem addressed is the lack of a comprehensive benchmark for evaluating decompilation technology, which is necessary for converting low-level binaries into human-readable source code. Previous efforts have relied on limited or synthetic benchmarks that do not accurately represent real-world binary-source mappings.

To address this issue, the authors created Decompile-Bench, which consists of two million binary-source function pairs generated from 100 million collected function pairs, totaling 450GB of binaries compiled from permissively licensed GitHub projects. The dataset is accompanied by a benchmark, Decompile-Bench-Eval, which includes manually crafted binaries from established datasets and compiled GitHub repositories released after 2025 to mitigate data leakage issues.

The results show that fine-tuning large language model-based decompilers with Decompile-Bench leads to a 20% improvement in re-executability rate compared to previous benchmarks. The authors also explored commonly used evaluation metrics to provide a thorough assessment of the studied decompilers. The code and data are publicly available on HuggingFace and GitHub, making it a valuable resource for advancing decompilation technology. Overall, Decompile-Bench provides a significant contribution to the field by offering a large-scale, real-world dataset for evaluating and improving decompilation accuracy.

📅 Published on May 19, 2025

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2505.12668
• PDF: https://arxiv.org/pdf/2505.12668

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#BinaryDecompilation #DecompilationBenchmarks #LargeLanguageModels #BinarySourceMapping #DecompilerEvaluation

🔥 VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillation

💡 The paper addresses the problem of multimodal on-policy distillation, where a student model learns from a privileged teacher model with fine-grained visual knowledge. However, the teacher's corrections are a mix of visual signals, linguistic priors, and teacher-specific effects, making it challenging to determine which corrections are supported by visual evidence. To tackle this issue, the authors introduce Visual Attribution Distillation, a counterfactual target reconstruction algorithm that estimates the visually attributable part of a teacher correction.

The method works by evaluating the teacher with and without relevant visual evidence at each student-generated prefix, and measuring the change in centered log probabilities. This defines a proxy for the visual evidence direction that estimates how revealing the evidence supports or refutes candidate tokens. The original correction is then projected onto this proxy to obtain an intervention-aligned component and a proxy-unexplained residual. The intervention-aligned component is used to reconstruct a student-anchored target, which serves as the primary supervision signal during training, while the privileged teacher contributes a weak regularizer.

The results show that Visual Attribution Distillation outperforms direct privileged-view distillation and visual-advantage weighting across six fine-grained visual benchmarks at 4B and 9B scales. Token-level and controlled-target analyses demonstrate that the proxy-aligned component is enriched in task-relevant visual corrections and yields stronger target shifts, especially when evidence refutes a mistaken answer. The findings support counterfactual target reconstruction as an effective alternative to source-mixed supervision, providing a more accurate and efficient way to transfer visual knowledge from the teacher to the student model.

📅 Published on Jul 30

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.28590
• PDF: https://arxiv.org/pdf/2607.28590

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#MultimodalOnPolicyDistillation #VisualAttributionDistillation #TargetReconstructionAlgorithms #MultimodalLearningMethods #OnPolicyDistillationTechniques

🔥 LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks

💡 The paper addresses the problem of long-horizon tasks that require sustained reasoning, tool use, and revision across many independent steps. Existing agent harnesses struggle to maintain task execution, task state, and completion assessment within a growing context, leading to difficulties in tracking the state and allowing incorrect self-assessments to propagate into later decisions.

To solve this problem, the authors reformulate long-horizon execution as a task-state management problem and propose LongHorizon-Harness. This approach maintains the task state explicitly outside of execution and updates it only with facts independently verified from the environment. The Manage-Execute-Audit loop uses a manager to maintain the task state and determine the next subtask, a fresh-context executor to perform it, and a read-only auditor to verify the resulting environment state before the next round.

The authors also introduce a lightweight Agent Adapter that supports interchangeable model and harness backends without modifying their native agent loops. The results show that LongHorizon-Harness improves performance on several benchmarks, including WaveBench, Terminal-Bench, and OSWorld, with significant gains across different models, harnesses, and interaction domains. For example, it improves Qwen from 51.8% to 80.7% on WaveBench, from 69.7% to 77.2% on Terminal-Bench, and from 2.8% to 8.3% on OSWorld. Additionally, it raises Claude Opus from 20.0% to 34.3% on an OSWorld subset, demonstrating consistent gains across different settings.

Overall, the paper contributes a novel approach to long-horizon task execution, providing a more effective and robust solution for real-world tasks that require sustained reasoning and decision-making.

📅 Published on Aug 3

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2608.01964
• PDF: https://arxiv.org/pdf/2608.01964
• Project Page: https://lh-harness.pages.dev

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#LongHorizonTasks #ArtificialIntelligenceAgents #TaskStateManagement #RealWorldApplications #AutonomousDecisionMaking

🔥 Orchard: An Open-Source Agentic Modeling Framework

💡 The paper introduces Orchard, an open source framework for scalable agentic modeling, which aims to transform large language models into autonomous agents capable of solving complex tasks. The problem addressed is that current research in agentic modeling is constrained by infrastructure and training gaps, with many high performing systems relying on proprietary codebases, models, or services. Most open source frameworks focus on orchestration and evaluation rather than scalable agent training.

The method presented is the Orchard framework, which consists of a lightweight environment service called Orchard Env, providing reusable primitives for sandbox lifecycle management across task domains, agent harnesses, and pipeline stages. On top of Orchard Env, three agentic modeling recipes are built: Orchard-SWE for coding agents, Orchard-GUI for vision language computer use agents, and Orchard-Claw for personal assistant agents.

The results show that Orchard achieves state of the art performance among open source models of comparable size. Specifically, Orchard-SWE achieves 64.3% and 67.5% on SWE-bench Verified after applying credit assignment and reinforcement learning. Orchard-GUI achieves 74.1%, 67.0%, and 64.0% success rates on WebVoyager, Online-Mind2Web, and DeepShop, respectively. Orchard-Claw achieves 59.6% pass@3 on Claw-Eval and 73.9% when paired with a stronger ZeroClaw harness.

The contributions of the paper are the introduction of the Orchard framework, which enables reusable agentic data, training recipes, and evaluations across domains, and the demonstration of its effectiveness in achieving state of the art performance in various tasks. The paper shows that a lightweight, open, harness agnostic environment layer can enable scalable agentic modeling, making it a significant contribution to the field of artificial intelligence.

📅 Published on May 14

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2605.15040
• PDF: https://arxiv.org/pdf/2605.15040

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#AgenticModeling #AutonomousAgents #ScalableAgentTraining #OpenSourceFrameworks #LargeLanguageModels

🔥 MLVC: Multi-platform Learned Video Codec for Real-World Deployment

💡 The paper introduces MLVC, a multi-platform learned video codec designed for practical cross-platform inference. Neural video codecs have surpassed classical codecs in coding efficiency but remain impractical for deployment due to cross-platform incompatibility and high computational cost. Existing quantization-based solutions fail to produce deterministic results across diverse hardware platforms, leading to catastrophic decoding failures.

The key idea of MLVC is to explicitly transmit scale parameters through the hyperprior, which guarantees entropy coding consistency across devices without requiring bit-exact arithmetic. While this increases bit rate overhead, the authors recover most of the coding efficiency through architectural improvements, including gated memory, ReGLU activation, a long-term reference recovery mechanism, and domain-specific perceptual training.

The results show that MLVC achieves over 70 percent BD-rate improvement over hardware HEVC, the strongest deployable baseline, while reaching subjective quality competitive with DVC-RT, which cannot operate across diverse platforms. Both the encoder and decoder run at 100 frames per second on average on commodity NPUs from Apple, Intel, and Qualcomm. MLVC is the first neural video codec to combine competitive compression performance, real-time speed, and cross-platform robustness across diverse consumer devices, making it suitable for widespread deployment.

📅 Published on Jun 26

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2606.28027
• PDF: https://arxiv.org/pdf/2606.28027

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#LearnedVideoCodecs #NeuralVideoCompression #CrossPlatformInference #VideoCodecDeployment #DeepLearningForVideoCoding