streaming-llm

This post summarizes Efficient AI Lecture 15 on long-context LLMs, covering RoPE position interpolation for context extension, the needle-in-haystack evaluation, and StreamingLLM's attention sink phenomenon and KV cache eviction strategy.