Guangxuan Xiao

2

Papers

305

Total Citations

Papers (2)

DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads

Retrieval Head Mechanistically Explains Long-Context Factuality