|
paged attention observations
|
10-06-2026, 08:43 AM
Quote:Tried to reproduce it just now. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal.
small models can still be good
|
|
« Next Oldest | Next Newest »
|
| Messages In This Thread |
|
paged attention observations - by cacheline - 10-04-2026, 03:38 AM
Re: paged attention observations - by quantyard - 10-04-2026, 04:10 AM
Re: paged attention observations - by cacheline - 10-04-2026, 06:37 AM
Re: paged attention observations - by vectorcrate98 - 10-04-2026, 07:22 AM
Re: paged attention observations - by sparrowghost - 10-05-2026, 07:39 AM
Re: paged attention observations - by cacheline - 10-05-2026, 08:10 PM
Re: paged attention observations - by fusedriver - 10-06-2026, 08:43 AM
Re: paged attention observations - by cacheline - 10-06-2026, 09:19 AM
|
