Kimi K2.5 Open-Sources Agent Swarm, CoT Plans Only 2-3 Steps Ahead

Featured

• Unified Personalized Reward Model for Vision Generation score 8
  入选 HF Daily Papers; HF 热度: 19 upvotes (+3); 关键词(7): DPO, GRPO, alignment, preference, reasoning
• How Well Do Models Follow Visual Instructions? VIBE: A Systematic Benchmark for Visual Instruction-Driven Image Editing score 8
  入选 HF Daily Papers; HF 热度: 16 upvotes (+3); 关键词(5): reasoning, multimodal, benchmark, evaluation, open-source
• Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics score 8
  入选 HF Daily Papers; HF 热度: 13 upvotes (+3); 关键词(2): fine-tuning, preference
• Enhancing Multi-Image Understanding through Delimiter Token Scaling score 10
  入选 HF Daily Papers; HF 热度: 5 upvotes (+2); 关键词(4): scaling, inference, vision-language, cost; 顶会接收: ICLR
• Kimi K2.5: Visual Agentic Intelligence score 9
  入选 HF Daily Papers; HF 热度: 206 upvotes (+4); 关键词(8): latency, pre-training, agent, agentic, coding
• Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models score 9
  入选 HF Daily Papers; HF 热度: 123 upvotes (+4); 关键词(4): multimodal, search, benchmark, evaluation
• CodeOCR: On the Effectiveness of Vision Language Models in Code Understanding score 9
  入选 HF Daily Papers; HF 热度: 91 upvotes (+4); 关键词(6): efficient, efficiency, compression, inference, multimodal
• Closing the Loop: Universal Repository Representation with RPG-Encoder score 9
  入选 HF Daily Papers; HF 热度: 81 upvotes (+4); 关键词(3): agents, reasoning, planning
• UniReason 1.0: A Unified Reasoning Framework for World Knowledge Aligned Image Generation and Editing score 9
  入选 HF Daily Papers; HF 热度: 74 upvotes (+4); 关键词(5): agent, reasoning, planning, multimodal, text-to-image
• No Global Plan in Chain-of-Thought: Uncover the Latent Planning Horizon of LLMs score 9
  入选 HF Daily Papers; HF 热度: 65 upvotes (+4); 关键词(2): reasoning, planning

Also Worth Noting

• Toward Enhancing Representation Learning in Federated Multi-Task Settings score 4
  关键词(1): efficient; 顶会接收: ICLR
• Moonworks Lunara Aesthetic II: An Image Variation Dataset score 4
  关键词(2): fine-tuning, evaluation; Reddit r/ML 热议
• FlyPrompt: Brain-Inspired Random-Expanded Routing with Temporal-Ensemble Experts for General Continual Learning score 4
  关键词(1): efficient; 顶会接收: ICLR
• Governance at the Edge of Architecture: Regulating NeuroAI and Neuromorphic Systems score 2
  关键词(4): efficiency, latency, edge, embodied
• Harnessing Flexible Spatial and Temporal Data Center Workloads for Grid Regulation Services score 2
  关键词(2): latency, real-time
• MarkCleaner: High-Fidelity Watermark Removal via Imperceptible Micro-Geometric Perturbation score 2
  关键词(4): efficient, inference, real-time, alignment
• RAPT: Model-Predictive Out-of-Distribution Detection and Failure Diagnosis for Sim-to-Real Humanoid Robots score 2
  关键词(3): lightweight, deployment, reasoning
• White-Box Neural Ensemble for Vehicular Plasticity: Quantifying the Efficiency Cost of Symbolic Auditability in Adaptive NMPC score 2
  关键词(3): efficiency, latency, cost
• Qrita: High-performance Top-k and Top-p Algorithm for GPUs using Pivot-based Truncation and Selection score 2
  关键词(4): efficient, throughput, search, evaluation
• When Is Rank-1 Enough? Geometry-Guided Initialization for Parameter-Efficient Fine-Tuning score 2
  关键词(5): efficient, fine-tuning, alignment, multimodal, vision-language
• A Relative-Budget Theory for Reinforcement Learning with Verifiable Rewards in Large Language Model Reasoning score 2
  关键词(3): efficient, efficiency, reasoning
• EchoJEPA: A Latent Predictive Foundation Model for Echocardiography score 2
  关键词(2): efficiency, pretraining
• Toward a Machine Bertin: Why Visualization Needs Design Principles for Machine Cognition score 2
  关键词(3): reasoning, vision-language, benchmark