投机执行快3倍，不丢token反更快

重点关注

• SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and Planning score 10
  入选 HF Daily Papers；HF 热度: 52 upvotes (+4)；有代码实现；关键词(6): lightweight, serving, latency, throughput, agentic
• DA-Flow: Degradation-Aware Optical Flow Estimation with Diffusion Models score 10
  入选 HF Daily Papers；HF 热度: 40 upvotes (+4)；有代码实现；关键词(1): compression
• EVA: Efficient Reinforcement Learning for End-to-End Video Agent score 10
  入选 HF Daily Papers；HF 热度: 34 upvotes (+4)；有代码实现；关键词(3): fine-tuning, GRPO, reasoning
• Rethinking Token-Level Policy Optimization for Multimodal Chain-of-Thought score 10
  入选 HF Daily Papers；HF 热度: 21 upvotes (+4)；有代码实现；关键词(3): GRPO, reasoning, vision-language
• WildWorld: A Large-Scale Dataset for Dynamic World Modeling with Actions and Explicit State toward Generative ARPG score 9
  入选 HF Daily Papers；HF 热度: 77 upvotes (+4)；有代码实现
• SIMART: Decomposing Monolithic Meshes into Sim-ready Articulated Assets via MLLM score 8
  入选 HF Daily Papers；HF 热度: 35 upvotes (+4)；关键词(1): embodied
• UniGRPO: Unified Policy Optimization for Reasoning-Driven Visual Generation score 8
  入选 HF Daily Papers；HF 热度: 30 upvotes (+4)；关键词(4): scaling, GRPO, post-training, reasoning
• VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions score 8
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(3): lightweight, reasoning, vision-language；顶会接收: CVPR
• RealMaster: Lifting Rendered Scenes into Photorealistic Video score 7
  入选 HF Daily Papers；HF 热度: 24 upvotes (+4)
• Reconstruction-Guided Slot Curriculum: Addressing Object Over-Fragmentation in Video Object-Centric Learning score 7
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现；关键词(1): edge

也值得关注

• UniFunc3D: Unified Active Spatial-Temporal Grounding for 3D Functionality Segmentation score 4
  入选 HF Daily Papers；关键词(1): reasoning
• GeoTikzBridge: Advancing Multimodal Code Generation for Geometric Perception and Reasoning score 4
  关键词(2): code generation, reasoning；顶会接收: CVPR
• Think 360°: Evaluating the Width-centric Reasoning Capability of MLLMs Beyond Depth score 4
  关键词(1): reasoning；顶会接收: CVPR
• SOUPLE: Enhancing Audio-Visual Localization and Segmentation with Learnable Prompt Contexts score 4
  关键词(1): pre-training；顶会接收: CVPR
• ENC-Bench: A Benchmark for Evaluating Multimodal Large Language Models in Electronic Navigational Chart Understanding score 4
  关键词(1): reasoning；顶会接收: CVPR
• Synthetic Mixed Training: Scaling Parametric Knowledge Acquisition Beyond RAG score 4
  机构: Stanford；关键词(3): scaling, RAG, synthetic data
• Cluster-Wise Spatio-Temporal Masking for Efficient Video-Language Pretraining score 4
  关键词(1): pretraining；顶会接收: CVPR
• Looking Beyond the Window: Global-Local Aligned CLIP for Training-free Open-Vocabulary Semantic Segmentation score 4
  关键词(1): scaling；顶会接收: CVPR
• DAK-UCB: Diversity-Aware Prompt Routing for LLMs and Generative Models score 4
  关键词(1): text-to-image；顶会接收: ICLR
• ViKey: Enhancing Temporal Understanding in Videos via Visual Prompting score 4
  机构: Carnegie Mellon；关键词(2): lightweight, reasoning
• PhysSkin: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning score 4
  关键词(1): real-time；顶会接收: CVPR
• SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Scheduling score 4
  机构: Microsoft Research；关键词(2): scaling, reasoning
• DetPO: In-Context Learning with Multi-Modal LLMs for Few-Shot Object Detection score 4
  机构: CMU；关键词(2): fine-tune, pre-training
• Improving LLM Predictions via Inter-Layer Structural Encoders score 3
  机构: Meta
• Who Spoke What When? Evaluating Spoken Language Models for Conversational ASR with Semantic and Overlap-Aware Metrics score 3
  机构: Carnegie Mellon