Hugging Face Daily Papers
500 articles archived · Visit source ↗ · RSS
-
Hugging Face Daily Papers research 1mo ago
SHAPE of Chain-of-Thought in Math Reasoning
Abstract SHAPE analyzes chain-of-thought reasoning via semantic spaces and heuristics to diagnose LLM mathematical reasoning and improve post-training. Generated by thinkingmachines/Inkling-Small Large language models (LLMs) achieve strong performance on mathematical reasoning…
7 -
Dynamic Important Example Mining for Reinforcement Finetuning
Abstract DIEM adaptively selects and reweights training examples during reinforcement fine-tuning by estimating each sample's marginal contribution to policy improvement and constraining batch updates to stabilize optimization. Generated by thinkingmachines/Inkling-Small…
21 -
Normalized Low-Rank Adaptation
Abstract Normalized Low-Rank Adaptation stabilizes LoRA training by normalizing down-projection matrices, accelerating convergence and improving performance without extra parameters or inference cost. Generated by thinkingmachines/Inkling-Small While low-rank adaptation (LoRA)…
26 -
Lucida: Parse, Generate, and Place for Composable Real-to-Sim Scene Modeling
Abstract Lucida improves composable indoor scene reconstruction by distributing pipeline requirements across parsing, asset generation, and VLM-guided placement to achieve high-fidelity editable replicas from cluttered captures. Generated by thinkingmachines/Inkling-Small…
37 -
CogEvol: Towards Efficient and Reliable Learning Environment Generation
Abstract CogEvol is a family of models that generate structured learning artifacts in a single pass using supervised fine-tuning and reinforcement learning with vision-language rewards, achieving high quality with far fewer parameters and lower cost. Generated by…
6 -
GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling
Abstract Direct end-to-end training of latent generative models avoids collapse via entropy preservation and asymmetric dynamics, using a generation-first strategy to achieve state-of-the-art image synthesis and unified multimodal generation. Generated by…
19 -
CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents
Abstract CAST improves LLM agent reliability by generating structured action-level rationales from sparse outcomes to train critique and policy models. Generated by thinkingmachines/Inkling-Small Large language model (LLM) agents are increasingly deployed in long-horizon,…
38 -
DreamX-Creator: Democratizing Native Audio-Video Generation at 2K Resolution
Abstract A compact 7B native joint audio-video generator uses cross-modal attention, progressive joint training, reinforcement learning with multimodal feedback, and an autoregressive 2K refinement pipeline to produce synchronized high-resolution outputs. Generated by…
30 -
MNIST-PRO: MNIST is Back as a Partially Observable World for AI Agents
Abstract The MNIST-PRO benchmark isolates perceptual-state construction in partially observable settings, revealing that multimodal agents struggle to integrate fragmented glimpses, continue exploring, and revise incorrect beliefs. Generated by thinkingmachines/Inkling-Small AI…
8 -
Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents
Abstract AutoSciRub improves autonomous scientific agents by generating task-specific executable rubrics that guide experiments, verify criteria, and iteratively refine outputs. Generated by thinkingmachines/Inkling-Small Autonomous scientific research agents are increasingly…
13 -
Scaling Large Reasoning Models beyond Human Supervision: A Path toward Superintelligence
Abstract This work proposes a structured ladder for scaling large reasoning models beyond human supervision by tracing autonomous rewards and self-generated experience, while identifying risks and evaluation dimensions. Generated by thinkingmachines/Inkling-Small Recent advances…
33 -
Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
Abstract On-policy distillation relies mainly on suppressing low-probability tokens rather than teacher guidance, motivating a supervision-free entropy-adaptive method that substantially improves reasoning performance. Generated by thinkingmachines/Inkling-Small On-policy…
29 -
PaperGym: Rubric-Centered Evolution for Research-Plan Generation
Abstract PaperGym converts scientific papers into training environments by separating research questions from evaluation rubrics, enabling reinforcement learning that improves research planning across multiple model sizes. Generated by thinkingmachines/Inkling-Small Research…
22 -
Cross-lingual Functional Vectors for Emotion Detection in Large Language Models
Abstract Function vectors extracted from one language improve multilingual emotion recognition by capturing language-independent task signals and reducing inference overhead. Generated by thinkingmachines/Inkling-Small Function vectors (FVs) have recently emerged as a promising…
37 -
Ask or Answer: A Decision Framework for Multi-Turn Health Misinformation Intervention
Abstract RO-PnR learns when to ask clarifying questions versus deliver corrections by weighing interaction costs against expected gains, improving cost-adjusted outcomes in health misinformation dialogues. Generated by thinkingmachines/Inkling-Small Correcting health…
21 -
Acquire, Repair, Preserve: A Diagnosis-Guided Post-Training Recipe for Small-Model Dialogue Game Agents
Abstract A 2B open-weight model improves interactive dialogue-game performance through supervised fine-tuning and turn-local preference training, though gains remain largely within the targeted game family. Generated by thinkingmachines/Inkling-Small Interactive dialogue games…
6 -
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
Abstract EvoUndo evaluates recoverability of self-modifying LLM agents and shows that reliable recovery requires co-designing verification, state grounding, and recovery-language expressivity. Generated by thinkingmachines/Inkling-Small LLM agents increasingly modify their own…
13 -
Generative Semantic Scene Completion
Abstract Generative discrete diffusion recasts outdoor LiDAR semantic scene completion via paired sparse-dense synthesis, semantic-guided generation, and structured refinement to achieve state-of-the-art results. Generated by thinkingmachines/Inkling-Small Outdoor LiDAR semantic…
28 -
Sliding-window beats linear attention
Abstract Sliding window attention with sinks outperforms post-trained linear attention on long-context tasks without requiring retraining, offering a cheaper and more reliable inference solution. Generated by thinkingmachines/Inkling-Small Due to the nature of quadratic…
5 -
Lost in Compression: A Controlled Cross-Lingual Audit of Extractive Prompt Compressors
Abstract Learned prompt compressors trained on English data disproportionately degrade non-English contexts, widening token-cost disparities, while multilingual training and deterministic methods reduce this gap. Generated by thinkingmachines/Inkling-Small Extractive prompt…
22 -
Act with Intent: Distilling Behavior Intent for Vision-Language-Action Models
Abstract Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the…
29 -
LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
Abstract LoopArena benchmarks how well a controller model guides a separate coding agent through long tasks, revealing low strict success rates and significant cost reductions. Generated by thinkingmachines/Inkling-Small Loop Engineering is emerging as a practice for organizing…
28 -
GGSS: Geodesic-Gated Spherical Steering for Inference-Time Debiasing of Generative Vision-Language Models
Abstract GGSS reduces demographic bias in generative vision-language models by steering visual tokens along geodesic arcs with an adaptive gate, preserving visual-language accuracy. Generated by thinkingmachines/Inkling-Small Generative vision-language models (VLMs) are…
19 -
Language Chain in Alignment: Cross-lingual Ranking Preference Optimization
Abstract Cross-lingual Ranking Preference Optimization improves multilingual alignment by transferring English preference knowledge to target languages through hierarchical ranking optimization. Generated by thinkingmachines/Inkling-Small The alignment of Large Language Models…
35 -
Rubric-to-Code Credit Assignment for Reinforcement Learning
Abstract RCCA improves reinforcement learning for interactive web app generation by assigning localized credit to code regions based on rubric-level feedback, yielding substantial benchmark gains. Generated by thinkingmachines/Inkling-Small Interactive web application generation…
29 -
Video Generative Models as Geometry Learner
Abstract GeoNeXt repurposes pretrained video generative models as a unified framework for geometry estimation via next-frame prediction, enabling efficient joint modeling of depth and surface normals with minimal labeled data. Generated by thinkingmachines/Inkling-Small Recent…
38 -
Agentic Artifact Creation: Systems, Evaluation, Principles, and Opportunities
Abstract This survey defines agentic artifact creation as stateful, feedback-driven construction of deliverables by AI systems and analyzes 259 works across artifact families, settings, and evaluation practices to propose principles for accountable control. Generated by…
35 -
DART-SD: Diamond-topology Aware Retrieval and Tuning for Self-Distillation of Multi-Turn Tool-Calling Agents
Abstract DART-SD improves multi-turn tool-calling agents by modeling execution as a diamond-topology graph, identifying critical failure points, and applying localized self-distillation to preserve valid reasoning while correcting errors. Generated by…
12 -
Locate Anything in Videos: Rethinking Efficient Generative Spatio-Temporal Video Grounding
Abstract Parallel Tube Decoding enables simultaneous spatial and temporal video grounding by removing autoregressive dependencies, drastically cutting latency while improving accuracy. Generated by thinkingmachines/Inkling-Small Spatio-temporal video grounding (STVG) requires…
37 -
Fast Weight Attention for Continual Learning
Abstract Recurrent fast-weight memories and selective state-space models are analyzed as online learning rules under autoregressive semantics, yielding normalized update families with stable renormalization that improve length extrapolation. Generated by…
27 -
Ring Forcing: Towards Precise Long-Term Memory for Autoregressive Video Diffusion
Abstract Ring Forcing is an autoregressive video diffusion framework that improves long-term memory through ring-structured training, history compression, and sparse rotary embeddings to achieve minute-long coherence. Generated by thinkingmachines/Inkling-Small Scaling video…
4 -
LayerRecall: A State-Conditioned Memory Router for Long-Horizon Consistency in Video Generation
Abstract LayerRecall selectively routes long-range historical memory into specific video diffusion layers to improve long-video consistency, supervised by cross-horizon prediction matching. Generated by thinkingmachines/Inkling-Small Autoregressive video diffusion enables…
32 -
Puro-2B: Poor Lab's Qwen2-1.5B Trained on RTX 5090 within $5090
Abstract A cost-efficient open-source pretraining recipe trains 2B-parameter models on consumer GPUs for under $7K, yielding performance near larger baselines while deriving cost scaling laws and studying data curricula. Generated by thinkingmachines/Inkling-Small Language model…
15 -
Paint What You See: Benchmarking Dexterous Visual Tool Use in Multimodal Agents
Abstract EASEL benchmarks fine-grained visual tool use through reference-guided reconstruction and semantic tasks, revealing that current multimodal agents struggle with closed-loop precision and trajectory stability. Generated by thinkingmachines/Inkling-Small Evaluation is…
23 -
StarHarness: Evolving Harnesses with Stratified Search for Enterprise Environments
Abstract StarHarness evolves fixed-weight agent harnesses via stratified task pools and hidden selection to improve enterprise tool-use performance and cross-model transfer. Generated by thinkingmachines/Inkling-Small We present StarHarness, a framework for evolving…
27 -
J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data
Abstract J-Zero enables self-improving language models across verifiable and unverifiable domains through adversarial co-evolution of a task generator, solver, and judge using predefined preference pairs. Generated by thinkingmachines/Inkling-Small Self-evolving language models…
32 -
Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction
Abstract ABot-Recon achieves stable long-horizon streaming 3D reconstruction by using only local temporal context and frame-independent predictions composed sequentially, reducing drift via a lightweight temporal refiner and composition-aware pose loss. Generated by…
36 -
StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing
Abstract StepGuard is a step-level guard model that audits agent actions before execution, trained via automatic trajectory generation and balanced reinforcement learning to reduce attacks with minimal utility loss. Generated by thinkingmachines/Inkling-Small LLM-based agents…
27 -
LMSM: LLM Security Framework Inspired by Linux Security Modules
Abstract LMSM applies a modular security framework to LLM serving by separating evidence calibration, policy evaluation, and output gating, enabling flexible interpretability-based enforcement without rebuilding request handling. Generated by thinkingmachines/Inkling-Small Large…
19 -
Training, learning and inference: unified dynamics of neural systems
Abstract Atomic generation facts compiled into a graph enable recursive scientific processes, and training-learning dynamics in nanoGPT are characterized by state-conditioned parameter updates, persistent functional reorganization, and frozen inference projections validated…
25 -
Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models
Abstract VLAct improves vision-language-action model performance by pre-training on diverse robot data with preserved vision-language priors and shared action semantics, achieving strong results across simulations and unseen embodiments with limited compute. Generated by…
32 -
PonderPounce: A Pretrained MLLM as an Episode Context Engine for Robot Control
Abstract PonderPounce leverages native causal context in multimodal language models as robot episode memory, jointly training a reasoning System2 module and a fast System1 action model to improve long-horizon policy performance without dedicated memory architectures. Generated…
14 -
Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning
Abstract Code-as-World represents physical environments as executable code to enable quantitative reasoning and scalable supervision for vision-language models. Generated by thinkingmachines/Inkling-Small Physical understanding and reasoning depend on forming compact and…
33 -
Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge
Abstract ElephantBench evaluates whether large language models retain multiple conflicting accounts of long-tail facts through a graph-based pipeline that creates verifiable multi-account questions, revealing widespread epistemic incompleteness. Generated by…
6 -
ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
Abstract ContextPilot improves long-horizon agent reasoning by expanding context-editing tools and using reinforcement learning with branch sampling to identify critical context decisions. Generated by thinkingmachines/Inkling-Small Long-horizon agentic tasks require large…
18 -
Luce: Relightable Gaussians for 3D Asset Generation
Abstract Luce unifies geometry and PBR materials in a voxelized Gaussian cloud, using a variational autoencoder and rectified-flow transformer to generate relightable 3D assets from single images. Generated by thinkingmachines/Inkling-Small High-fidelity image-to-3D generation…
23 -
CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes
Abstract CritICL improves LLM reasoning at inference time by using structured failure patterns from weaker models as critique-based guidance, reducing generation and token costs. Generated by thinkingmachines/Inkling-Small Recent advances in inference-time scaling have…
9 -
What Does an Evaluation License? A Commit-Bound Census of Claim-Relative Inference in Inspect Evals
Abstract Evaluation artifacts specify a forward computation: a task, scorer, and reported metric. They do not necessarily license the claim attached to that metric because the historical evidence and alternative semantics needed to replay it may be unbound. We formalize this…
25 -
EditaLive! Unified Character Video Editing for Live Streaming
Abstract EditaLive enables real-time human-centric live-stream video editing by adapting an image animation model to causal streaming generation with distilled two-step sampling and sparse attention. Generated by thinkingmachines/Inkling-Small Conventional video editing…
10 -
Magpie: Real-Time World Renderer for Interactive Games
Abstract Magpie is a real-time generative rendering system that separates gameplay logic from visual generation to preserve interactive designability while reducing asset requirements for game prototypes. Generated by thinkingmachines/Inkling-Small Modern game development relies…
24