Pruning a Small Model Is a Shortcut Only on a Tight Budget

Featured

• ClinHallu: A Benchmark for Diagnosing Stage-Wise Hallucinations in Medical MLLM Reasoning score 9
  机构: Alibaba; 入选 HF Daily Papers; 有代码实现; 关键词(2): fine-tuning, reasoning
• OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence Chains score 8
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 有代码实现; 关键词(2): fine-tuning, reasoning
• VISTA: View-Consistent Self-Verified Training for GUI Grounding score 8
  入选 HF Daily Papers; HF 热度: 5 upvotes (+2); 有代码实现; 关键词(1): GRPO
• From Chatbot to Digital Colleague: The Paradigm Shift Toward Persistent Autonomous AI score 7
  入选 HF Daily Papers; HF 热度: 16 upvotes (+3); 关键词(2): tool use, reasoning

Also Worth Noting

• Small LLMs: Pruning vs. Training from Scratch score 4
  机构: Princeton; 关键词(1): pruning
• DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation score 4
  关键词(2): real-time, retrieval-augmented; 顶会接收: ICML
• OdysSim: Building Foundation Models for Human Behavior Simulation score 4
  机构: CMU; 关键词(2): distillation, post-training
• From Prompts to Responses: Dual-Sided Data Leakage and Defense in Split Large Language Models score 4
  关键词(2): deployment, fine-tuning; 顶会接收: ICML
• SkillAudit: Ground-Truth-Free Skill Evolution via Paired Trajectory Auditing score 4
  机构: Huawei; 关键词(2): deployment, edge
• HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry score 4
  入选 HF Daily Papers; 关键词(1): scaling
• One Layer's Trash is Another Layer's Treasure: Adaptive Layer-wise Visual Token Selection in LVLMs score 4
  关键词(5): lightweight, compression, pruning, deployment, vision-language; 顶会接收: CVPR
• ForceForget: Reinforcement Concept Removal for Enhancing Safety in Text-to-Image Models score 4
  关键词(1): text-to-image; 顶会接收: ICML
• Hy-Embodied-0.5-VLA: From Vision-Language-Action Models to a Real-World Robot Learning Stack score 4
  入选 HF Daily Papers; 关键词(6): deployment, fine-tuning, post-training, pre-training, vision-language
• Fodor and Pylyshyn's Systematicity Challenge Still Stands score 3
  机构: MIT