Geometry Conflict Predicts Continual Fine-Tuning Forgetting

Featured

• Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training score 10
  入选 HF Daily Papers; HF 热度: 24 upvotes (+4); 有代码实现; 关键词(2): fine-tuning, post-training
• Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction score 9
  入选 HF Daily Papers; HF 热度: 10 upvotes (+3); 有代码实现; 关键词(4): lightweight, compression, reasoning, vision-language
• Metal-Sci: A Scientific Compute Benchmark for Evolutionary LLM Kernel Search on Apple Silicon score 7
  入选 HF Daily Papers; HF 热度: 3 upvotes (+1); 有代码实现; 关键词(1): lightweight
• Shaping Schema via Language Representation as the Next Frontier for LLM Intelligence Expanding score 6
  入选 HF Daily Papers; HF 热度: 5 upvotes (+2); 关键词(1): scaling
• Crosslingual On-Policy Self-Distillation for Multilingual Reasoning score 6
  入选 HF Daily Papers; 有代码实现; 关键词(4): scaling, distillation, GRPO, reasoning
• Reinforcing Multimodal Reasoning Against Visual Degradation score 5
  入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 关键词(4): compression, fine-tuning, GRPO, reasoning
• DeltaRubric: Generative Multimodal Reward Modeling via Joint Planning and Verification score 5
  入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 关键词(1): reasoning

Also Worth Noting

• SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent Learning score 4
  入选 HF Daily Papers; 关键词(3): coding, embodied, open-source
• Learning-Augmented Scalable Linear Assignment Problem Optimization via Neural Dual Warm-Starts score 4
  关键词(1): lightweight; 顶会接收: ICML
• FedCIGAR: A Personalized Reconstruction Approach for Federated Graph-level Anomaly Detection score 4
  关键词(1): synthetic data; 顶会接收: IJCAI
• Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairs score 4
  关键词(2): DPO, text-to-image; 顶会接收: ICML
• Don't Click That: Teaching Web Agents to Resist Deceptive Interfaces score 4
  关键词(2): deployment, vision-language; 顶会接收: ACL
• Any2Any 3D Diffusion Models with Knowledge Transfer: A Radiotherapy Planning Study score 4
  关键词(1): post-training; 顶会接收: CVPR
• Learning Multi-Indicator Weights for Data Selection: A Joint Task-Model Adaptation Framework with Efficient Proxies score 4
  关键词(3): fine-tuning, instruction tuning, reasoning; 顶会接收: IJCAI
• MOTOR-Bench: A Real-world Dataset and Multi-agent Framework for Zero-shot Human Mental State Understanding score 4
  关键词(1): reasoning; 顶会接收: CVPR
• Entropy-informed Decoding: Adaptive Information-Driven Branching score 4
  关键词(2): code generation, reasoning; 顶会接收: ICML
• CrossVL: Complexity-Aware Feature Routing and Paired Curriculum for Cross-View Vision-Language Detection score 4
  关键词(1): vision-language; 顶会接收: CVPR
• Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models score 3
  入选 HF Daily Papers
• Dystruct: Dynamically Structured Diffusion Language Model Decoding via Bayesian Inference score 3
  入选 HF Daily Papers
• Outlier-Robust Diffusion Solvers for Inverse Problems score 3
  顶会接收: CVPR