500AI
Back to researchers

Guangxuan Xiao

Streaming + long-context stability (attention sinks)

Co-authored Attention Sinks: a practical trick for stable streaming and long-context attention.

Highlights

Long contextInferenceEfficiency
Focus: Streaming + long-context stability (attention sinks)
Why it matters: Co-authored Attention Sinks: a practical trick for stable streaming and long-context attention.
Start here
Efficient Streaming Language Models with Attention Sinkspaper →

Research Areas

Long contextInferenceEfficiency
Guangxuan Xiao - AI Researcher Profile | 500AI