Synthetic long-context tasks at controllable lengths that reveal where effective context starts to break down.
Read the original source — github.com
benchmark · Shared by tscosj
0 comments
No comments yet.
RULER
Synthetic long-context tasks at controllable lengths that reveal where effective context starts to break down.