Cunxiao Du

Muon Outperforms Adam in Tail-End Associative Memory Learning

Anatomy of a Hybrid Mind: Deconstructing Hybrid Reasoning in Large Language Models

When Attention Sink Emerges in Language Models: An Empirical View

Efficient Inference for Large Language Model-based Generative Recommendation

SWIFT: On-the-Fly Self-Speculative Decoding for LLM Inference Acceleration

SimLayerKV: A Simple Framework for Layer-Level KV Cache Reduction

BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms