Xiaoran Liu

MrRoPE: Mixed-radix Rotary Position Embedding

Beyond Real: Imaginary Extension of Rotary Position Embeddings for Long-Context LLMs

SIM-CoT: Supervised Implicit Chain-of-Thought

ProtoKV: Long-context Knowledges Are Already Well-Organized Before Your Query

Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache

VideoRoPE: What Makes for Good Video Rotary Position Embedding?

ReAttention: Training-Free Infinite Context with Finite Attention Scope

LongSafetyBench: Long-Context LLMs Struggle with Safety Issues