You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
i think it might be a good idea to make the LLM max length a bit longer or more like let the LLM finish before it stops because half the LLM answer is missing or its incomplete most times and sometimes it reguires several dozen calls to the LLM until it gives a completed Answer and it should be changed to something like 750-1000 Max lenght(Tokens) maybe even higher
The text was updated successfully, but these errors were encountered:
okay, got it. now is 5k, and also top_k and top_p next version.
when u click generate button; it will send into SD, do u really want send more then 1000 token into SD clip encoder?
update...
after i try for FLUX. hnn, so many token bring so many details.
yep, we need .
i did some testing myself in SD forge using Flux i got this result with 500 Max length Tokens and 20 top k and 0.9 top p and 0.7 LLM temperature
more max length tokens might be better to be used with flux
i think it might be a good idea to make the LLM max length a bit longer or more like let the LLM finish before it stops because half the LLM answer is missing or its incomplete most times and sometimes it reguires several dozen calls to the LLM until it gives a completed Answer and it should be changed to something like 750-1000 Max lenght(Tokens) maybe even higher

The text was updated successfully, but these errors were encountered: