Repositories Become Skills as Models Manage Their Own Context

Featured

• Language Models Can Control Their Own Attention score 11
  机构: MIT; 入选 HF Daily Papers; HF 热度: 60 upvotes (+4); 关键词(1): lightweight
• Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills score 10
  入选 HF Daily Papers; HF 热度: 519 upvotes (+4); 有代码实现; 关键词(1): distillation
• EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction score 10
  入选 HF Daily Papers; HF 热度: 113 upvotes (+4); 有代码实现; 关键词(3): lightweight, distillation, agentic
• PaperCompiler: Faithful Paper-to-Code Generation via Repository-Level Specification Compilation score 8
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 有代码实现; 关键词(2): code generation, coding
• Cliff: Learning Process Rewards from the First Mistake score 7
  入选 HF Daily Papers; HF 热度: 16 upvotes (+3); 关键词(4): distillation, GRPO, post-training, reasoning
• CRISP: Cliff-awaRe Input-adaptive Sparse Prefilling with Structural-Mass-Motivated Routing score 9
  入选 HF Daily Papers; HF 热度: 8 upvotes (+2); 关键词(1): real-time; 顶会接收: EMNLP
• Post-Training Language Models for Gold-Medal Performance in Coding Competitions score 6
  入选 HF Daily Papers; HF 热度: 9 upvotes (+2); 关键词(4): fine-tuning, post-training, coding, reasoning
• Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens score 6
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1); 有代码实现
• Debias-SparseGPT: Bias-Aware Pruning for Large Language Models score 8
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1); 关键词(5): compression, quantization, pruning, deployment, post-training; 顶会接收: EMNLP

Also Worth Noting

• Belief-Calibrated Optimization: An Explicit World Model for Agentic Optimization score 4
  机构: Amazon; 关键词(3): agentic, coding, reasoning
• READY or Not: Reliable Enterprise Agent Deployment score 4
  机构: Carnegie Mellon; 关键词(1): deployment
• A Unified Rate-Distortion Perspective on Vector, Product, and Scalar Quantization score 4
  机构: University of Toronto; 关键词(3): compression, quantization, coding
• C$^{3}$T: Counterfactual Causal Reasoning for Sentiment Shifts in Social-Media Conversation Trees score 4
  机构: Google; 关键词(1): reasoning
• User Feedback Provides a Unique Signal that LLMs Can not Detect score 4
  机构: MIT; 关键词(1): synthetic data
• Entangled Representations Amplify Collateral Damage in Unlearning score 3
  机构: University of Toronto
• DiffIE: Diffusion-based Open Information Extraction score 3
  机构: Huawei