δ-mem用8×8矩阵换长上下文

重点关注

• $\delta$-mem: Efficient Online Memory for Large Language Models score 10
  入选 HF Daily Papers；HF 热度: 101 upvotes (+4)；有代码实现；关键词(2): lightweight, fine-tuning
• SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture score 10
  入选 HF Daily Papers；HF 热度: 158 upvotes (+4)；有代码实现；关键词(4): post-training, agentic, reasoning, vision-language
• Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions score 7
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现
• Mela: Test-Time Memory Consolidation based on Transformation Hypothesis score 7
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；有代码实现
• LychSim: A Controllable and Interactive Simulation Framework for Vision Research score 7
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现；关键词(5): serving, pretraining, agentic, reasoning, synthetic data
• DeepRefine: Agent-Compiled Knowledge Refinement via Reinforcement Learning score 7
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(1): reasoning
• AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive score 7
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(2): agentic, reasoning
• CausalCine: Real-Time Autoregressive Generation for Multi-Shot Video Narratives score 8
  入选 HF Daily Papers；HF 热度: 24 upvotes (+4)；关键词(1): real-time
• NanoResearch: Co-Evolving Skills, Memory, and Policy for Personalized Research Automation score 8
  入选 HF Daily Papers；HF 热度: 10 upvotes (+3)；有代码实现
• LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues score 7
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现；关键词(3): latency, RAG, coding

也值得关注

• A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models score 5
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)
• jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers score 5
  入选 HF Daily Papers；HF 热度: 7 upvotes (+2)
• PlantMarkerBench: A Multi-Species Benchmark for Evidence-Grounded Plant Marker Reasoning score 4
  入选 HF Daily Papers；关键词(2): throughput, reasoning
• Phoenix-VL 1.5 Medium Technical Report score 4
  机构: Mistral；关键词(3): post-training, pretraining, data curation
• A Research Agenda on Agents and Software Engineering: Outcomes from the Rio A2SE Seminar score 4
  机构: Google；关键词(1): agentic
• AccLock: Unlocking Identity with Heartbeat Using In-Ear Accelerometers score 4
  机构: Peking University；关键词(1): deployment
• Rethink the Role of Neural Decoders in Quantum Error Correction score 4
  关键词(5): compression, quantization, deployment, latency, real-time；顶会接收: ICML
• Images in Sentences: Scaling Interleaved Instructions for Unified Visual Generation score 5
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；关键词(1): scaling