TokDrift: When LLM Speaks in Subwords but Code Speaks in Grammar
Paper
•
2510.14972
•
Published
•
34
Wow
-
-
LightMem: Lightweight and Efficient Memory-Augmented Generation
Paper • 2510.18866 • Published • 111 -
Every Attention Matters: An Efficient Hybrid Architecture for Long-Context Reasoning
Paper • 2510.19338 • Published • 114 -
The Smol Training Playbook
📚2.79kThe secrets to building world-class LLMs
-
Stabilizing Reinforcement Learning with LLMs: Formulation and Practices
Paper • 2512.01374 • Published • 96