120B on One GPU, and 40% of Video Benchmarks Are Guessable

Featured

• ClawsBench: Evaluating Capability and Safety of LLM Productivity Agents in Simulated Workspaces score 11
  机构: Apple; 入选 HF Daily Papers; HF 热度: 16 upvotes (+3); 有代码实现
• Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding score 10
  入选 HF Daily Papers; HF 热度: 201 upvotes (+4); 有代码实现; 关键词(2): reasoning, leaderboard
• Watch Before You Answer: Learning from Visually Grounded Post-Training score 10
  入选 HF Daily Papers; HF 热度: 26 upvotes (+4); 有代码实现; 关键词(4): post-training, reasoning, vision-language, data curation
• MegaTrain: Full Precision Training of 100B+ Parameter Large Language Models on a Single GPU score 10
  入选 HF Daily Papers; HF 热度: 25 upvotes (+4); 有代码实现; 关键词(1): throughput
• General Multimodal Protein Design Enables DNA-Encoding of Chemistry score 10
  入选 HF Daily Papers; HF 热度: 21 upvotes (+4); 有代码实现; 关键词(1): scaling
• MedGemma 1.5 Technical Report score 6
  入选 HF Daily Papers; HF 热度: 9 upvotes (+2); 关键词(1): reasoning

Also Worth Noting

• Improving Sparse Memory Finetuning score 4
  机构: Carnegie Mellon; 关键词(2): finetuning, open-source
• PaperOrchestra: A Multi-Agent Framework for Automated AI Research Paper Writing score 3
  机构: Yale
• Memory Dial: A Training Framework for Controllable Memorization in Language Models score 3
  顶会接收: ACL
• Multilingual Language Models Encode Script Over Linguistic Structure score 3
  顶会接收: ACL
• XMark: Reliable Multi-Bit Watermarking for LLM-Generated Texts score 3
  顶会接收: ACL