We make Chain-of-Thought reasoning in large language models (1) more efficient by creating implicit reasoning with lightweight language models; (2) still effective as the implicit reasoning maintains semantic alignment with ground-truth reasoning.

Theoretical analysis of scheduling algorithms for LLM queries with latency constraints when using RadixAttention along with a novel scheduling algorithm.

We propose two scalable DP algorithms for high-dimensional sparse variable selection, leveraging modern mixed-integer programming techniques.