TokDrift: When LLM Speaks in Subwords but Code Speaks in Grammar
Paper
ā¢
2510.14972
ā¢
Published
ā¢
35
Wow
-
-
LightMem: Lightweight and Efficient Memory-Augmented Generation
Paper ā¢ 2510.18866 ā¢ Published ā¢ 113 -
Every Attention Matters: An Efficient Hybrid Architecture for Long-Context Reasoning
Paper ā¢ 2510.19338 ā¢ Published ā¢ 115 -
The Smol Training Playbook
š2.96kThe secrets to building world-class LLMs
-
Stabilizing Reinforcement Learning with LLMs: Formulation and Practices
Paper ā¢ 2512.01374 ā¢ Published ā¢ 104