#Repetition Problem
1 messages · Page 1 of 1 (latest)
what model do you use.
Gpt-4, not Gpt-4o that was ended but regular Gpt-4
We don't support any openai models, only PAI and Cosmos that were made by Pawan. You could always try a jailbreak or adjusting the temperature but any repetition etc isn't anything we can really help with outside of that.
I’m using Cosmos as I have the pro-it v1 link in the Open AI Reverse Proxy box on Chub under Secrets but in Configuration/Presets, I have the Prompt Structure with the API being Open AI to match the Secrets with the model being Gpt-4.
Oh, model on websites dont change anything.
Only cosmo and PAI do, provided by pawan.
Those on the site in the box?
Dont matter
What are your settings on Chub that you use for it?
Oh. But do I change it? I can’t remove the Open AI parts on Chub.
My generation parameters?
It won't affect the model at all, it's just a box to fill. Because the sites are set up to use any proxy as an open ai proxy to identifiy, but it doesn't do anything to the model itself
Yeah Generation, temperature, etc
Oh, ok
I have temperature on 0.75, repetition penalty on 1.33, frequency penalty on 1.33, top p on 1.00, and top k on 1. I don’t know if I need to list max new token and context size, but those are my settings under the generation parameters in the configuration/presets on Chub.
Also, the preset settings is still under open ai when I said the temperature and other stuff
Maybe I could change it but I don’t know about switching the Open Ai to Kobold or Anthropic or the other ones.
Switching the model name won't change anything for the model.
Have you tried other bots and get the same issue?
Yes.
Hmn, everything otherwise looks good, I personally use a higher temperature myself, arounf the 1.0 and 1.2 range without issue, you could try raising it and see if it changes anything. Or switch sites and try it on some bots there and see if it pops up as an issue as well
And okay, I’ll pick the Anthropic model on Chub
Also, I don’t know about 1.0 and 1.2 as those would be too high and might have some silly answers. Before I used 0.75, I used to be having 0.40 or something like that but I’m used to 0.75 now and I haven’t tried 0.80 or 0.85 but 0.75 is not too high or low.
So I’ll stick with 0.75
Definitely up to your preference for sure 🙂 Pai was made for the low ranges, Cosmos is actually made for higher ranges, but it all depends on the responses you like. I find they get more creative and detailed when I'm on 1.2, without the Shakespeare other models are prone to
Okay. So I think my problem is fixed now as I changed to Palm2 actually but Palm2 wanted me to provide a key and I had to get one from Google. And I tried both models of Palm2 with Bison being the better half of the other one, even though it might have some issues.
I am honestly not sure what any of that is. It sounds like a different api altogether if you got a key from somewhere that wasn't here
And if so please refrain from speaking about it, as it can be considered advertising in the server, which may get mod-flagged
...
For how long has this been happening.
Well, I still have the api from here but Palm2 wasn’t taking it and I needed to get the api for it sadly.
And I did this in Chub.
But okay, I didn’t mean to advertise.
Oh yea, also.
#📢𝐀𝐧𝐧𝐨𝐮𝐧𝐜𝐞𝐦𝐞𝐧𝐭 message
Palm2 has nothing to do with cosmos... That's where my confusion lies. Cosmos is it's own api and the api keys come from this server (though with free version no key is needed, but as you are a supporter I'm guessing you are using Pro, or were using pro)
Very possible this is a server issue.
Oh. And I am using pro but I have both boxes of the Open Ai Reverse Proxy having the link from here and the Palm 2 key having the private key I put there for Palm under secrets in Chub.
So I don’t know what to tell ya’ll
What cosmos link are you using?
Because this is the way chub would be set up if using cosmos (screenshots provided) #🙌𝐒𝐞𝐭𝐮𝐩-𝐓𝐮𝐭𝐨𝐫𝐢𝐚𝐥 message
Except the link would be the pro link and the key would be the bot key from here.
No other key would work with cosmos other than what you get from the server
Unless you were using the free version, as it would take anything typed in the key box at that point
Maybe if you posted a creenshot of your settings it would be a bit clearer, because yes, it's difficult to help with the information given, as none of it makes sense with how the api works, so difficult to see what happened verses what should happen
Oh. And I said before that I’m using the pro-it v1 link, so I’ll take a screenshot and show you.
Both boxes for the things I said are filled
But I couldn’t chat without the Palm 2 Key box being empty
Ok I see what is happening, you have the info for cosmos but you aren't actually using cosmos. You're using palm2 which is a different api altogether supported by chub
You have both boxes filled but you don't actually need both. You don't have an api key in the reverse proxy so it couldn't use cosmos
Oh.
This is where you would need to put your api key if you are using the supporter version of cosmos
And you would get that from https://discord.com/channels/1055397662976905229/1064807234825113621 and doing a /key command in the channel there
Well, I guess I’ll have to put one there but it’ll take a bit of my 500 credits.
Yeah cosmos does use credits but they refresh daily, and it's only 1 credit per 10k tokens. SO unless you are doing a heck of a lot of writing for the entire day you don't really run out
You can test and keep track of that by doing a /info command in bot channel
As is right now the way you have it set up, it's not using credits because it's not using cosmos
So if Cosmos is the api you want to use, that's the way you set it up 🙂
But is Bison under Palm2 good, or nah? I’m testing it for now before I put in my api key for cosmos.
The other one that’s not Bison under Palm2 has the ‘Assistant’ thing, which I don’t like.
I can't speak for other apis not given here, only the ones that are provided by this server
Thats a big no no
Let us know if you continue to have trouble!