Hierarchical Sparse Attention Done Right: Toward Infinite Context Modeling
Paper • 2607.02980 • Published • 81
AI Research Papers/Things
A collection of research regarding AI that interests me.
-
-
Gemma 4 Technical Report
Paper • 2607.02770 • Published • 74 -
SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibe
Paper • 2607.03451 • Published • 33 -
TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training
Paper • 2607.05804 • Published • 18 -
MentalThink: Shaping Thoughts in Mental SVG World
Paper • 2607.03530 • Published • 14 -
Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding
Paper • 2607.05722 • Published • 13 -
TREK: Distill to Explore, Reinforce to Refine
Paper • 2607.05339 • Published • 8 -
Quantifying and Expanding the Theoretical Capacity of Late-Interaction Retrieval Models
Paper • 2607.05803 • Published • 10 -
PluraMath: Extending Mathematical Reasoning Evaluation Beyond High-Resource Languages
Paper • 2607.05992 • Published • 6 -
Attending to Multimodal Generation One Token at a Time
Paper • 2607.03738 • Published • 10 -
Accurate, Interdisciplinary and Transparent Structure-property Understanding with Deep Native Structural Reasoning
Paper • 2607.07708 • Published • 88 -
Automating the Design of Embodied Agent Architectures
Paper • 2606.30111 • Published • 13 -
Teaching LLMs a Low-Resource Language: Enhancing Code Completion in Pharo
Paper • 2607.04939 • Published • 5 -
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning
Paper • 2607.07508 • Published • 26 -
Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity
Paper • 2607.07386 • Published • 12 -
The Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement Learning
Paper • 2606.29526 • Published • 170 -
OmniOpt: Taxonomy, Geometry, and Benchmarking of Modern Optimizers
Paper • 2607.04033 • Published • 76 -
ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference Outcomes
Paper • 2607.04439 • Published • 63 -
MANCE: Manifold Aware Concept Erasure
Paper • 2607.03973 • Published • 49 -
Program-as-Weights: A Programming Paradigm for Fuzzy Functions
Paper • 2607.02512 • Published • 238 -
Dockerless: Environment-Free Program Verifier for Coding Agents
Paper • 2606.28436 • Published • 113 -
DOPD: Dual On-policy Distillation
Paper • 2606.30626 • Published • 113 -
BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding
Paper • 2606.31315 • Published • 77 -
UI-MOPD: Multi-Platform On-Policy Distillation for Continual GUI Agent Learning
Paper • 2607.04425 • Published • 72 -
Agents' Last Exam
Paper • 2606.05405 • Published • 385 -
Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs
Paper • 2605.30611 • Published • 252 -
On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters
Paper • 2606.02437 • Published • 241 -
MemSlides: A Hierarchical Memory Driven Agent Framework for Personalized Slide Generation with Multi-turn Local Revision
Paper • 2606.17162 • Published • 177 -
LoopCoder-v2: Only Loop Once for Efficient Test-Time Computation Scaling
Paper • 2606.18023 • Published • 210 -
MiniMax Sparse Attention
Paper • 2606.13392 • Published • 153 -
Domino: Decoupling Causal Modeling from Autoregressive Drafting in Speculative Decoding
Paper • 2605.29707 • Published • 152 -
Agentic Abstention: Do Agents Know When to Stop Instead of Act?
Paper • 2606.28733 • Published • 150 -
EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environments
Paper • 2606.13681 • Published • 143 -
Are We Ready For An Agent-Native Memory System?
Paper • 2606.24775 • Published • 130 -
Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
Paper • 2606.11926 • Published • 130 -
COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation
Paper • 2605.31264 • Published • 124 -
VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models
Paper • 2606.16140 • Published • 124 -
SWE-Explore: Benchmarking How Coding Agents Explore Repositories
Paper • 2606.07297 • Published • 123 -
SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning
Paper • 2606.13673 • Published • 111 -
OCC-RAG: Optimal Cognitive Core for Faithful Question Answering
Paper • 2606.00683 • Published • 102 -
Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Lineage-Grounded Idea Generation
Paper • 2607.08758 • Published • 41 -
Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition
Paper • 2601.16211 • Published • 54 -
UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks
Paper • 2607.08768 • Published • 34 -
Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE
Paper • 2607.07740 • Published • 24 -
Linear Attention Architectures: Mechanisms, Trade-offs, and Cross-Layer Routing
Paper • 2607.07953 • Published • 16 -
UP: Unbounded Positive Asymmetric Optimization for Breaking the Exploration-Stability Dilemma
Paper • 2607.06987 • Published • 9 -
Flash-BoN: Instant Drafts for Inference-Time Scaling in Diffusion Models
Paper • 2607.04461 • Published • 11 -
Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents
Paper • 2607.08716 • Published • 15 -
SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs
Paper • 2503.16163 • Published -
ShortGPT: Layers in Large Language Models are More Redundant Than You Expect
Paper • 2403.03853 • Published • 65 -
MoE-SpAc: Efficient MoE Inference Based on Speculative Activation Utility in Heterogeneous Edge Scenarios
Paper • 2603.09983 • Published • 3 -
R-KV: Redundancy-aware KV Cache Compression for Reasoning Models
Paper • 2505.24133 • Published • 2 -
Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading
Paper • 2607.08964 • Published • 76 -
Scalable Visual Pretraining for Language Intelligence
Paper • 2607.09657 • Published • 57 -
Trust Region Policy Distillation
Paper • 2607.04751 • Published • 34 -
KronQ: LLM Quantization via Kronecker-Factored Hessian
Paper • 2607.07964 • Published • 32 -
Self-Guided Test-Time Training for Long-Context LLMs
Paper • 2607.09415 • Published • 19 -
Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning
Paper • 2607.08393 • Published • 18 -
Weak-to-Strong Generalization via Direct On-Policy Distillation
Paper • 2607.05394 • Published • 138 -
ABot-AgentOS: A General Robotic Agent OS with Lifelong Multi-modal Memory
Paper • 2607.10350 • Published • 85 -
LoMo: Local Modality Substitution for Deeper Vision-Language Fusion
Paper • 2605.30265 • Published • 23 -
LeVLJEPA: End-to-End Vision-Language Pretraining Without Negatives
Paper • 2607.00784 • Published • 2 -
ViQ: Text-Aligned Visual Quantized Representations at Any Resolution
Paper • 2606.27313 • Published • 38 -
LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning
Paper • 2605.22012 • Published • 46 -
Latent Noise Mask for Reducing Visual Redundancy in Multimodal Large Language Models
Paper • 2606.30168 • Published -
LightSTAR: Efficient Visual Document Retrieval via Lightweight Selection with Vision-Adaptive Refinement
Paper • 2606.23539 • Published -
Vision Inference Former: Sustaining Visual Consistency in Multimodal Large Language Models
Paper • 2605.18160 • Published -
AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification
Paper • 2607.11849 • Published • 33 -
Metacognition in LLMs: Foundations, Progress, and Opportunities
Paper • 2607.11881 • Published • 29 -
DaisyChain-Web
🕸11Train a shared model P2P by opening a browser tab
-
Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation
Paper • 2607.11886 • Published • 84 -
Blind-Spots-Bench: Evaluating Blind Spots in Multimodal Models
Paper • 2607.08317 • Published • 35 -
Know Before Fix: QA-Driven Repository Knowledge Acquisition for Software Issue Resolution
Paper • 2607.11111 • Published • 23 -
MuScriptor: An Open Model for Multi-Instrument Music Transcription
Paper • 2607.08168 • Published • 21 -
Principled Analysis of Deep Reinforcement Learning Evaluation and Design Paradigms
Paper • 2607.07769 • Published • 10 -
What LLM Forecasters Know but Don't Say: Probing Internal Representations for Calibration and Faithfulness
Paper • 2607.08046 • Published • 14 -
TernaryLLM: Ternarized Large Language Model
Paper • 2406.07177 • Published • 1 -
TernaryLM: Memory-Efficient Language Modeling via Native 1-Bit Quantization with Adaptive Layer-wise Scaling
Paper • 2602.07374 • Published • 2 -
Ternary MobileNets via Per-Layer Hybrid Filter Banks
Paper • 1911.01028 • Published -
Binary and Ternary Natural Language Generation
Paper • 2306.01841 • Published • 3 -
Tequila: Trapping-free Ternary Quantization for Large Language Models
Paper • 2509.23809 • Published • 4 -
Sherry: Hardware-Efficient 1.25-Bit Ternary Quantization via Fine-grained Sparsification
Paper • 2601.07892 • Published • 4 -
TerDiT: Ternary Diffusion Models with Transformers
Paper • 2405.14854 • Published • 2 -
Harness Handbook: Making Evolving Agent Harnesses Readable,Navigable, and Editable
Paper • 2607.13285 • Published • 227 -
Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning
Paper • 2607.12395 • Published • 99 -
KnowAct-GUIClaw: Know Deeply, Act Perfectly, Personal GUI Assistant with Self-Evolving Memory and Skill
Paper • 2607.12625 • Published • 80 -
SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration
Paper • 2607.15257 • Published • 71 -
SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning
Paper • 2607.14777 • Published • 103 -
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget
Paper • 2607.14952 • Published • 203 -
Spectral Rewiring for Exploration, Purification, and Model Merging
Paper • 2607.03065 • Published • 25 -
MeanFlowNFT: Bringing Forward-Process RL to Average-Velocity Generators
Paper • 2607.15273 • Published • 17 -
DeepLoop: Depth Scaling for Looped Transformers
Paper • 2607.13491 • Published • 16 -
GRASP: GRanularity-Aware Search Policy for Agentic RAG
Paper • 2607.10463 • Published • 9 -
Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models
Paper • 2607.15277 • Published • 9 -
RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM
Paper • 2607.11683 • Published • 146 -
RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources
Paper • 2606.29538 • Published • 141 -
Loop the Loopies!
Paper • 2607.16051 • Published • 73 -
xHC: Expanded Hyper-Connections
Paper • 2607.14530 • Published • 56 -
RecGPT-V3 Technical Report
Paper • 2607.15591 • Published • 30 -
On-Policy Delta Distillation
Paper • 2607.15161 • Published • 38 -
From Human-Centric to Agentic Code Review: The Impact of Different Generations of Generative AI Technology on Review Quality
Paper • 2607.13196 • Published • 29 -
Understanding Reasoning from Pretraining to Post-Training
Paper • 2607.16097 • Published • 28 -
Recursive Harness Self-Improvement
Paper • 2607.15524 • Published • 24 -
Beyond Entropy: Correctness-Aware Advantage Shaping via Contrastive Policy Optimization
Paper • 2607.14614 • Published • 12 -
DSWorld: A Data Science World Model for Efficient Autonomous Agents
Paper • 2607.15901 • Published • 12 -
SWE-Pruner Pro: The Coder LLM Already Knows What to Prune
Paper • 2607.18213 • Published • 77 -
Apple-π: Benchmarking Thinking with Video Towards Law-Grounded Physical Intelligence
Paper • 2607.16401 • Published • 43 -
Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints
Paper • 2607.18144 • Published • 15 -
Group Entropy-Controlled Policy Optimization
Paper • 2607.16850 • Published • 29 -
Environment-free Synthetic Data Generation for API-Calling Agents
Paper • 2607.16900 • Published • 20 -
Text Template Tokens Are Implicit Semantic Registers in Diffusion Transformers
Paper • 2607.19139 • Published • 73 -
DataFlow-Harness: A Grounded Code-Agent Platform for Constructing Editable LLM Data Pipelines
Paper • 2607.16617 • Published • 137 -
Stale but Stable: Staleness-Adaptive Trust Regions for Stabilizing Asynchronous Reinforcement Learning
Paper • 2607.18722 • Published • 35 -
Subliminal Clocks: Latent Time Modelling in Diffusion Language Models
Paper • 2607.01774 • Published • 38 -
SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD
Paper • 2607.20145 • Published • 63 -
Beyond Relevance-Centric Retrieval: Rubric-Oriented Document Set Selection and Ranking
Paper • 2607.19747 • Published • 31 -
Scaling Laws for Hypernetwork-Based Knowledge Injection in Large Language Models
Paper • 2607.19604 • Published • 16 -
SLPO: Scaling Latent Reasoning via a Surrogate Policy
Paper • 2607.19691 • Published • 5 -
AREX: Towards a Recursively Self-Improving Agent for Deep Research
Paper • 2607.21461 • Published • 147 -
K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs
Paper • 2605.09635 • Published • 59 -
Tencent WorkBuddy Bench: A Multi-Domain Coding-Agent Benchmark with Contamination-Resistant Task Construction
Paper • 2607.20911 • Published • 25 -
NVIDIA-labs OO Agents: Native Python Object-Oriented Agents
Paper • 2607.20709 • Published • 31 -
Molt: A Scalable PyTorch-Native Training Framework for Agentic Reinforcement Learning
Paper • 2607.21653 • Published • 29 -
Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills
Paper • 2607.22529 • Published • 35 -
DataPrep-Bench: Benchmarking LLMs as Training Data Preparators
Paper • 2607.20465 • Published • 49 -
Three-Body Scattering for Generative Modeling
Paper • 2607.18198 • Published • 13 -
Scaling Native Multimodal Pre-Training From Scratch
Paper • 2607.22043 • Published • 22 -
Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture Problems
Paper • 2607.21503 • Published • 21 -
LAMAR: An Open Language-Aware Multilingual Alignment Reranker
Paper • 2607.22042 • Published • 11 -
IDEAgent: Agentic Quality-Diversity Search for Research Idea Generation
Paper • 2607.22375 • Published • 7 -
Multi-Head Latent Control: A Unified Interface for LLM Agent Decision Making
Paper • 2607.14277 • Published • 8 -
From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search
Paper • 2607.24280 • Published • 67 -
Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
Paper • 2607.24731 • Published • 63 -
LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V6-GGUF
Image-Text-to-Text • 35B • Updated • 99.7k • 196