#Questions/Issues
1 messages Β· Page 1 of 1 (latest)
hey @narrow epoch could you send the urls as links so they show up as blue?
thanks
I can't embed images, but, the AI is spewing out a bunch of random stuff
Chinese characters on my side
Yeah, same here.
Same here, Iβll try to get a screenshot of it
No response from API
Same for me
Same bro
odd.
Same here. First one was good, then it started doing this
Maybe it respond so quickly that it canβt fixed the errors?
is it fixed now?
yup, same here hpmie
yeah, it works now
Yes
Damn this is fast
Ok itβs better now, no more random Chinese letters
Yeah, it is very speedy
Agree, feels weird not having to wait a minute for it to generate
Relatively long answers too, at least at the beginning of a new chat
It's faster, but with some answers, like: ??? LOL
Seems to respect the * * for inner monologue
Definitely an improvement in the formatting of the text
Empty response from API, I think the problem is me, try to see my configs
holy shit, i got like 3 whole paragraphs in 5 seconds
Oh yeah, at least from the beginning of new chats I can see a clear improvement on the bot personality. It actually focuses on how the bot is feeling - and what they're thinking.
Wow, that's really fast
holy mother of god this is fast
yeah
Exactly, same feedback
People, remember to post feedback in the testing channel with the feedback template, thank you π
Thanks
Im entering the api but i keep getting a error that saids it invalid
create a help thread https://discord.com/channels/1055397662976905229/1122465350110412841
Also do look on the common issues thread.
same, but when i use 8k it works
Make sure it's set up correctly
it is
it twlls me to get help from this server
then make a post in https://discord.com/channels/1055397662976905229/1122465350110412841
"Error during generation, error: Error: Please wait a few minutes then try again. if the problem persists, please contact us on discord. https://discord.pawan.krd"
As I said, make a post in help
Well then.
Wait.
π
It's working just great for me tbh fast responses no lagging and everythings on point
Use it for a bit, then give your feedback.
On silly tavern unfortunately error when connecting....
try using on venus chub :) (tip: doesn't work)
Put in the first link to the model that work for me on ST
Works fine
ik that the model link must be the first
Not for me.
for ST, chub yes
2, links work fine on jai
I deleted the promts but left the persona in the ST , did the gender and pronouns catch on , I wondered
fixed
Already?
open ai configs should not be used, use chub's configs instead
thats the problem of everything.
Wdym chub's configs?
Well-
Okay?
preset settings, use chub's and you are ready to go
I have openAI selected and it's working fine π€·ββοΈ
Anyone with issues should create a help thread instead of asking for help here.
https://discord.com/channels/1055397662976905229/1122465350110412841
for me it breaks
i wonder if this cosmo is pro or not
if its pro then we will probably have 10k on normal one
This 32k is definitely impressive.
In the beginning if anypov and speak in the first person will be on they/them, although it takes time to start talking trying to go to the third then again on the first, in theory, in old chats if I write in the first person already remembers pronouns.
Especially having just been brought in.
If you write a response to a certain number of tokens for example 250 in the old chat may give out a similar number of answers. While trying, I'll try to find large texts in chat and copy them, while I have tokens worth 600. So I'll try it out on the new chats
Pawan, what are the recommended generation settings to use on chub ai this model v3?
set temperature to 0.7 and that's all, other settings are already set from our server.
hm, that's what I did but the messages I receive are good but they don't respect the uses of grammar. The commas, the quotation marks. I don't know what I did π€
try it with a few different bots, and then when you finished testing you can give your feedback about the issue if you saw the same issue with all the bots
Okie dokie! Tysm
Bless Pawan, he's giving me some motivation to actually make heavy-lore bots with his incredible models. β€οΈ
Well, NSFW is a bit of an almost-pressure. Still testing. The answers are varied and not repetitive. That's good.
what about pacing?
it varies depending on the character and the nature of the character in the story with us. I'm trying one who's already obsessed with the character, so it'll be faster.
You have to user cold bots, depending on the situation. Try it out.
I'm still having the issue that says that I need to wait.
How long will the test be available?
+12 more hours
Thx
Periodically the IP is lost
What?
Nevermind, I've already gone to Help.
I'm Sorry I was blind for moment there, I couldn't not find the Public Early Testing, I moved it to this one. π
Having an issue where it says "A network error occurred, you may be rate limited or having connection issues: Failed to fetch (unk)"
It also gives me failed to fetch
same
Same.
i have a question, how do i use "openAI default generation settings" in agnaistic?
it gives me a 502 bad gateway error
For those that have already tested this version the other day (early testers), do we need to give feedback as well or no? Since we've already done so prior to the public test.
((Also having the same problem as Mooth, "A network error occurred, you may be rate limited or having connection issues: Failed to fetch (unk)" ))
We are all having our dogs failing to fetch. I wish I could have tried it more before falling asleep when the β hit 1000
Failed to fetch !
someone help me with this pls
Hi, I just got the error "Network error. Try again later!", everything was working a couple of hours ago. What could be the problem?
Servers appear to be down right now.
me as well, what is going on?
As I said.
The test does not appear to be working right now.
This is Questions/Issues about Early-Testing the new Model, not about ANY website.
I wonder if this is a fair question then. With being a supporter, do the credits you have add overtime, or must all be used in one day, and with your preset settings and how often you use CosmosRP 32k, are you satisfied with your daily useage?
Credits reset every day, the unused credits don't stack.
Glad @stray ore got my back, cause I have no idea either lmao.
I never check my credits, cause since I'm supporter 1, I get unlimted credits.
on CosmoRp-Pro
The credits go a looooong way. Even without supporter 1 it's basically unlimited on the 16k wit 1 credit using 10,000 tokens. I never run out lol
Unsure what the credit situation will be like with 32k
I did skim over an unlimited option, but all I see on Patreon for 1 is you have 750 tokens per day xD
You have to look at the model page hereon discord to see how many tokens it equals to, it's different for every model
750 credits = 7,500,000 tokens basically for 16k pro a day
With cosmorp-PRO, supporter I and up have unlimited access
or actually unlimited because 750 is supp 1
so yeah doesn't even take credits on pro
For every model besides Pro I see for being "Public". Does that mean there is a way to gain tokens outside of being a supporter?
But to have unlimited on 16k seems really good and to use the credits per day for 32k.
Using Chub, wish there was multiple instances of OpenAI reverse proxie so you could easily switch from one to the other when the other can't be used for the day. That's the only reason I see credits to be accumulative rather then reset every day to be a good thing. Perhaps Accumulative only to be completely reset the start of every month would be good? Many things to think about, but I only barley touched the 32k to consider it xD
Yes, everyone in the server as a non supporter automatically gets 250 credits per day. You have them as well, you can check by doing /info in the bot channel. If you boost the server, you also get an additional 100 per day. If you are active, you get 1 wallet credit per 10 messages, which acts like daily credits however those are one use and don't refresh daily
Being a supporter gives more credits + unlocks the ability to use the supporter only models
Never realized that. Gonna have to test it out when I can.
yall servers will be back up soon anw so it's chill
@grand ore 1-day timeout, i hope this is enough time to learn reading
@meager scaffold you too
It won't let me post my message in the early-testing due to it being a review of the nsfw side of things (I did spoiler it but... y'know.)
I can try and uh... imply things? but that feels wrong to do given that'd be trying to dodge the censors.
Might have to keep it more general. "NSFW was rushed in some more active instances" or "when it came to harder NSFW scenes" blah blah blah (not a real feedback from me, just examples of how I've had to do it in the past)
yeah, I already tried to generalize it somewhat while still giving clear feedback but
Oof, what's pinging you, auto?
I put it bluntly in some places so now I'm doing the french thing
(Oil of the nut of the coco)
yeahhh
The bot overlord is cranky today π¦
Ah, its not working for me again. RIP. Servers must be overloaded.
Neither for me
Fixed
Got this error: PROXY ERROR 400: {"status":false,"error":{"message":"Please wait a few minutes then try again. if the problem persists, please contact us on discord. https://discord.pawan.krd","type":"invalid_request_error"}} (unk)
Ignore me, its fine again. Thanks π
Getting a network error.
Same.
mine just types an asterisk and then fails to finish its response every time
Unable to send messages through
After all, it's a test, mistakes and workload is inevitable given how much people test
Best to inform since it is a test, but you're right
my messages aren't getting through....
stuck replying
proxy error unknown object
link unk
PROXY ERROR: Unknown response: [object Object] (unk)
there we go
The server periodically stops due to certain issues, but it has been fixed.
for now
Once again
yes. I came because the message doesn't load
For free? I don't think so. I believe you have to be a supporter.
Im sure pawan is working very hard on it.
Error in Input stream unk
fixed, again
Thank you for all the effort Pawan, that's gotta be frustrating lol
Has to be
I have some strange problem and I have no idea what to do(((
This is a pain for you Pawan, but I thank you for the trouble.
im sorry pawan
lmao
I try to do NSFW in a test model and the bot stops...every 1 -2 responses, probably a sign of something.
bonk
double bonk
stop be ho*ny. yea
π π π»
the bot is offended
I'm still having the issue that says I have to wait.
Is the bot still up?
naur
nops
@narrow epoch extend the test period, the model wasn't working the entire night ππ
bruh
It wasn't for me since yesterday. ):
the model works perfectly fine for me
There were some
Large periods of time when it was down
I suppose due to instability, since it is testing and a lot of people must be trying it
Server down again I believe
I just got up to it again, only got a few messages in before it went down again π
fixed
It's working! Thank you!
Down!
This? Error during generation, error: Error: Please wait a few minutes then try again. if the problem persists, please contact us on discord.
im a fibber ignore that
Still stuck for me
It keeps appearing to me.
It's back!
Really?
Was for like 2 responses at least, but yeah nvm down again
Did it die again? π
same
PROXY ERROR: No response from bot (pgshag2)
@here The test server will be down for 10 mins, i'm trying fo fix this issue
Can't blame you, that model is sick asf
probably yeah
the model is good tho i think its worth the wait
@here I deployed a temporary fix to the test server. Now, if the server stops, it will automatically restart itself. So, if you encounter an error, wait 10-15 seconds and try again.
Is anyone else's bots devolving into caveman speak after each restart? The quality just went way down on my chat from earlier π I'm only noticing it just before I get a network message that signals the server is down, and it starts dropping articles like "the" "a" "is" etc which makes it sound like a caveman describing something.
Mine seems fine, though it has started to stop using pronouns like 'I' and "You"
The measages are failing
Same for mine, they keep cutting off or not working
Works alright for me
They do occasionally cut off or give an error but again, that is to be expected
Given that this is just testing
what will happen after the public testing ends? π€
My guess is the 16k context Cosmosrp will be the unlimited one and the new 32k V3 one will be only for supporters. Itβs just a guess though.
Anyone else has a problem with the 32k version writing very shakespearean-like dialogues? I've been tinkering with the temp and it seems nothing is working
Happens
I looked at the feedbacks and none mentioned about the writing being a little shakespearean. It feels unnatural and robotic
Yes, it has a bit of that language sometimes among other things. I was going to update my feedback with more things to add, like that for example, but my feedback is so high up in the chat that it wouldn't be helpful.
felt
I haven't gotten that, exactly, but I have gotten a few where it somewhat ignores the personality
Or it'll repeat a phrase (regenerating or swiping tends to fix that)
Occasionally, a continue will cause it to repeat the whole message again
And it does not correctly put dots, commas and names. Things like that
I would like to try more and see but it crashes frequently (which I understand)
@narrow epoch One question sir. The instructions are clear, Jailbreak is not allowed, but what about Main Prompts?
Technically their function is identical, the difference is that Jailbreaks is a constant reminder message, and the Main Prompt is a single instruction at the beginning of the chat. Unless otherwise configured.
@severe crane any instruction or prompt to change the response behavior is not recommended.
Ah. I'm incredibly sorry, I wish I hadn't missed that.
Any prompt that changes the model's response counts as an instruction, even if it's part of the character's description text. Some creators add jailbreaks to the character's description .
All janitor bots basically. Luckily I use my own bots without any of those but almost all of them have that inside
It's okay; it's necessary for private tests, but in public tests, it's not a big problem.
Okay, good to know. Either way, I'm super excited to get my hands back on this, thanks for all the hard work developing and working through this all!
I wonder when It'll come out officially :O
I hope soon.
What counts as a promt that changes the model's behaivor?
I too hope soon.
A question from before since I've been thinking about it, but what time does this wallet system reset each day? Or is that accumulative?
Accumulative, you dont lose them until you use them.
The daily credits reset every day, and you can do a /info to see what time in your timezone
Thanks. by the looks of it I may be a supporter when the 32k version comes out. I really liked the improvements xD
What's your opinion between that and the 16k?
I honestly really love the 16k and use it all the time, it's been great. THough when the 32k comes out I will absolutely switch to it, it feels even more of a step up to the 16k one, details wise and I already thought 16k was pretty detailed to begin with lol
Damn, here I'm hoping that it'll also be free to supports, which is why I wanted to ask about the wallet system just in case.
But, all in all, it preformed exceptionally well. I tested it with a bot designed to be a robot manufacturing itself, Kept personality traits and all the fun gizmos I added on to it till the end.
And-
Preformed quite well even without me constantly reminding the bot of what it had, creatively used what I gave it without being needed to be reminded, even if I did that once or twice to make sure.
It pops me something like that on Silly Tavern from time to time: "Chat Completion API
Please wait a few minutes then try again. if the problem persists, please contact us on discord. https://discord.pawan.krd"
Is that because server is busy?
The testing is finished.
If you got questions or issues with the other models, make a help thread