4B Agent on 10K Data, MoE Upcycling Saves 32% Compute

Featured

• Expert Upcycling: Shifting the Compute-Efficient Frontier of Mixture-of-Experts score 9
  入选 HF Daily Papers; HF 热度: 15 upvotes (+3); 有代码实现; 关键词(3): scaling, pre-training, MoE
• DR-Venus: Towards Frontier Edge-Scale Deep Research Agents with Only 10K Open Data score 10
  入选 HF Daily Papers; HF 热度: 45 upvotes (+4); 有代码实现; 关键词(6): scaling, deployment, latency, edge, fine-tuning
• SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks score 9
  入选 HF Daily Papers; HF 热度: 14 upvotes (+3); 有代码实现; 关键词(2): scaling, open-source
• SWE-chat: Coding Agent Interactions From Real Users in the Wild score 7
  入选 HF Daily Papers; HF 热度: 10 upvotes (+3); 关键词(2): coding, open-source
• Convergent Evolution: How Different Language Models Learn Similar Number Representations score 5
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2)

Also Worth Noting

• CreativeGame:Toward Mechanic-Aware Creative Game Generation score 5
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1); 关键词(1): code generation
• Surrogate modeling for interpreting black-box LLMs in medical predictions score 3
  机构: Harvard