my "universal" preset - temp 0.87, top_p 0.81, rep_pen 1.03 - works quite well with 2.2
#Sao10K/L3-70B-Euryale-v2.1
fiery oxide
iron drift
I use it through Infermatic
I have the infer sub, use it on SillyTavern with the instruct prompt from Sao and recommended settings
The only thing I changed is rep pen to 1.08 since 1.1 makes Euryale skip possessive pronouns for some reason
Yea when I use it on ST it's definitely good. Less horny than 2.1 and each swipe is still very unique. I hope it works out for you!
wraith birch
https://huggingface.co/Sao10K/L3.3-70B-Euryale-v2.3
New Euryale is out, so I hope it will be added to openrouter as well
lone cave
Maybe if deepinfra hosting it we can get it throught OR?
#1070782759716520038 message
I heard the new llama is better overall base for fine tune, so it's should be intresting to see that model on OR.
warm iris
it seems the context for l3.3-euryale-70b on infermatic is 32k but on openrouter it is 8k. could you increase it to match the provider?
daring shale
it seems the context for l3.3-euryale-70b on infermatic is 32k but on openrouter...
We set it like that on purpose 
warm iris
too bad, 8k is not really enough
daring shale
😞 I know, we will look if it’s feasible to increase it
real marsh
https://openrouter.ai/sao10k/l3.3-euryale-70b
6 month later
oh its a different model
oops
daring shale
we have the two versions hosted now, v2.2 and v2.3
wet pelican
We set it like that on purpose <:7847gothkitty:1215308304557936690>
Why though?
daring shale
Why though?
It's what we can offer for now 😦
wet pelican
No, I'm not complaining. I just wanna know why.
There's usually some technical reason right?
(Could it be that you have to reach a certain tier in order to be able to send in that many tokens?)
daring shale
No, I'm not complaining. I just wanna know why.
There's usually some technical ...
With popular models increase the context slows the model in general, we are trying to keep the balance with this model
wet pelican
Slows the model for everyone because the providers load it onto a given server and larger contexts can cause it to have that slowdown right?
Would that mean that in the future, when there is less traffic, you would go back and increase context size again? Right?
daring shale
Less traffic or more resources, correct
daring shale
too bad, 8k is not really enough
We are working on this now, it'll be 16K @subtle phoenix ☝️
daring shale
For all the ones interested, it's now 16K
https://openrouter.ai/sao10k/l3.3-euryale-70b
wet pelican
