they still can't fix the 65k token barrier, we've made no progress since 2021
| https://i.imgur.com/YbMiOBE.png | 09/13/26 |
Poast new message in this thread
Date: September 13th, 2026 5:42 PM
Author: https://i.imgur.com/YbMiOBE.png
yeah I run 232k+ context windows, but only to catch overruns. I set compression threshold to 50% so that I don't blow past 128k tokens under normal use. At 65k tokens it starts to slow down, and by 128k it's forgetting important details. There's no model that has ever solved this problem.
(http://www.autoadmit.com/thread.php?thread_id=5903592&forum_id=2most#50134007) |
|
|