VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training
Paper • 2602.10693 • Published • 220
to read
-
-
Flash-KMeans: Fast and Memory-Efficient Exact K-Means
Paper • 2603.09229 • Published • 82 -
DIVE: Scaling Diversity in Agentic Task Synthesis for Generalizable Tool Use
Paper • 2603.11076 • Published • 5 -
LongCat-Flash-Prover: Advancing Native Formal Reasoning via Agentic Tool-Integrated Reinforcement Learning
Paper • 2603.21065 • Published • 77 -
2Xplat: Two Experts Are Better Than One Generalist
Paper • 2603.21064 • Published • 25 -
Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?
Paper • 2603.24472 • Published • 53 -
The Y-Combinator for LLMs: Solving Long-Context Rot with λ-Calculus
Paper • 2603.20105 • Published • 37 -
TAPS: Task Aware Proposal Distributions for Speculative Sampling
Paper • 2603.27027 • Published • 141 -
DreamLite: A Lightweight On-Device Unified Model for Image Generation and Editing
Paper • 2603.28713 • Published • 19 -
Think Anywhere in Code Generation
Paper • 2603.29957 • Published • 25 -
Embarrassingly Simple Self-Distillation Improves Code Generation
Paper • 2604.01193 • Published • 34 -
The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook
Paper • 2604.02029 • Published • 137 -
Therefore I am. I Think
Paper • 2604.01202 • Published • 30 -
Beyond Accuracy: Unveiling Inefficiency Patterns in Tool-Integrated Reasoning
Paper • 2604.05404 • Published • 40 -
Self-Execution Simulation Improves Coding Models
Paper • 2604.03253 • Published • 32 -
RAGEN-2: Reasoning Collapse in Agentic RL
Paper • 2604.06268 • Published • 55