|
How do rolling buffers preserve initial sink tokens and recent local context? |
Answered by
Ultron09
Aug 28, 2026
Replies: 1 comment
|
StreamingLLM retains initial 4 sink tokens and slides a circular FIFO window for recent sequence positions. |
0 replies
Answer selected by
Ultron09
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
StreamingLLM retains initial 4 sink tokens and slides a circular FIFO window for recent sequence positions.