编码答案而非问题，embedding涨9%

重点关注

• LLM2Vec-Gen: Generative Embeddings from Large Language Models score 13
  机构: Mila；入选 HF Daily Papers；HF 热度: 28 upvotes (+4)；有代码实现；关键词(2): distillation, reasoning
• Prism-$Δ$: Differential Subspace Steering for Prompt Highlighting in Large Language Models score 8
  入选 HF Daily Papers；HF 热度: 10 upvotes (+3)；有代码实现
• CodePercept: Code-Grounded Visual STEM Perception for MLLMs score 8
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现；关键词(4): scaling, distillation, code generation, reasoning
• Are Video Reasoning Models Ready to Go Outside? score 8
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；有代码实现；关键词(5): deployment, reasoning, vision-language, embodied, open-source
• COMIC: Agentic Sketch Comedy Generation score 8
  机构: University of Washington；入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(2): production, agentic
• Geometric Autoencoder for Diffusion Models score 7
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(1): compression
• Neural Field Thermal Tomography: A Differentiable Physics Framework for Non-Destructive Evaluation score 7
  机构: Princeton；入选 HF Daily Papers；关键词(1): synthetic data
• Dynamics-Predictive Sampling for Active RL Finetuning of Large Reasoning Models score 7
  机构: Tsinghua；关键词(2): finetuning, reasoning；顶会接收: ICLR
• Learning to Assist: Physics-Grounded Human-Human Control via Multi-Agent Reinforcement Learning score 7
  机构: Carnegie Mellon；关键词(2): real-time, robotics；顶会接收: CVPR
• $V_{0.5}$: Generalist Value Model as a Prior for Sparse RL Rollouts score 6
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)；关键词(3): real-time, GRPO, reasoning

也值得关注

• UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations score 4
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)
• Federated Active Learning Under Extreme Non-IID and Global Class Imbalance score 4
  关键词(1): lightweight；顶会接收: CVPR
• The Curse and Blessing of Mean Bias in FP4-Quantized LLM Training score 4
  入选 HF Daily Papers；关键词(1): quantization
• DSFlash: Comprehensive Panoptic Scene Graph Generation in Realtime score 4
  关键词(6): deployment, latency, edge, fine-tune, reasoning；顶会接收: CVPR
• Recover to Predict: Progressive Retrospective Learning for Variable-Length Trajectory Prediction score 4
  关键词(1): distillation；顶会接收: CVPR
• Guiding Diffusion Models with Semantically Degraded Conditions score 4
  关键词(2): lightweight, text-to-image；顶会接收: CVPR
• Bilevel Layer-Positioning LoRA for Real Image Dehazing score 4
  关键词(1): fine-tuning；顶会接收: CVPR
• When Fine-Tuning Fails and when it Generalises: Role of Data Diversity and Mixed Training in LLM-based TTS score 4
  机构: Amazon；关键词(3): latency, fine-tuning, finetuning
• GLM-OCR Technical Report score 4
  机构: Zhipu；关键词(4): deployment, production, throughput, edge
• RCTs & Human Uplift Studies: Methodological Challenges and Practical Solutions for Frontier AI Evaluation score 4
  机构: Carnegie Mellon；关键词(1): deployment
• Mind the Sim2Real Gap in User Simulation for Agentic Tasks score 4
  机构: Carnegie Mellon；关键词(3): serving, agentic, open-source
• Reactive Writers: How Co-Writing with AI Changes How We Engage with Ideas score 3
  机构: Microsoft Research
• Verbalizing LLM's Higher-order Uncertainty via Imprecise Probabilities score 3
  机构: Oxford
• On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD score 3
  顶会接收: AAAI
• Muscle Synergy Priors Enhance Biomechanical Fidelity in Predictive Musculoskeletal Locomotion Simulation score 3
  机构: Carnegie Mellon