Two Loops Take SWE-bench From 43 to 64

Featured

• OPD-Evolver: Cultivating Holistic Agent Evolver via On-Policy Distillation score 10
  入选 HF Daily Papers; HF 热度: 26 upvotes (+4); 有代码实现; 关键词(1): distillation
• GameCraft-Bench: Can Agents Build Playable Games End-to-End in a Real Game Engine? score 10
  入选 HF Daily Papers; HF 热度: 42 upvotes (+4); 有代码实现; 关键词(1): coding
• Variable-Width Transformers score 8
  入选 HF Daily Papers; HF 热度: 5 upvotes (+2); 有代码实现; 关键词(2): scaling, MoE
• Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification score 9
  入选 HF Daily Papers; HF 热度: 11 upvotes (+3); 有代码实现; 关键词(4): scaling, quantization, fine-tuning, pre-training
• ActWorld: From Explorable to Interactive World Model via Action-Aware Memory score 6
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 关键词(3): compression, real-time, reasoning
• Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients score 8
  入选 HF Daily Papers; HF 热度: 48 upvotes (+4); 关键词(3): distillation, GRPO, vision-language

Also Worth Noting

• FoundCause: Causal Discovery with Latent Confounders from Observational Data score 4
  机构: Amazon; 关键词(3): edge, reasoning, synthetic data
• SuCo: Sufficiency-guided Continuous Adaptive Reasoning score 4
  关键词(2): fine-tuning, reasoning; 顶会接收: ICML
• Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Models score 4
  机构: Huawei; 关键词(3): GRPO, post-training, reasoning
• SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs score 4
  关键词(2): scaling, MoE; 顶会接收: ICML
• LoopCoder-v2: Only Loop Once for Efficient Test-Time Computation Scaling score 8
  入选 HF Daily Papers; HF 热度: 134 upvotes (+4); 关键词(6): scaling, latency, instruction tuning, agentic, code generation
• Looped World Models score 6
  入选 HF Daily Papers; HF 热度: 8 upvotes (+2); 关键词(1): scaling
• The Discrete-Log Clock: How a Transformer Learns Modular Multiplication score 3
  机构: Stanford
• Temporal Preference Optimization for Unsupervised Retrieval score 3
  顶会接收: ICML
• Bridging Functional Correctness and Runtime Efficiency Gaps in LLM-Based Code Translation score 3
  顶会接收: ICML
• MLLMs Get It Right, Then Get It Wrong: Tracing and Correcting Late-Layer Textual Bias score 3
  顶会接收: IJCAI
• Adaptive Volumetric Mechanical Property Fields Invariant to Resolution score 3
  机构: NVIDIA