/wAIfu/ AI Vtuber Chatbots

[Anonymous No.]

A thread dedicated to the discussion of AI Vtuber Chatbots.

>No bitrate Biboo edition

/wAIfu/ Status: Rock hard

>/wAIfu/ Chatbot list:
[/wAIfu/ Bot List]https://rentry.org/wAIfu_Bot_List_Final
[4chan Bot List]https://rentry.org/meta_bot_list

>/wAIfu/ Tavern Card Archive:
https://mega.nz/folder/cLkFBAqB#uPCwSIuIVECSogtW8acoaw
[Tavern Card Editor]https://zoltanai.github.io/character-editor/
[Easily Port CAI bots to Tavern Cards]https://rentry.org/Easily_Port_CAI_Bots_to_tavern_cards

>Tavern/Character Ai (Character Creation) Guides:
[CAI Guides and Text formatting Scripts]https://rentry.org/CAI_Guides_and_Text_Formatting_Scripts
[Tavern - basic]https://rentry.org/alichat
[Tavern - basic]https://rentry.org/botmaking_tips
[Tavern - advanced]https://rentry.org/MothsBotMakingStuff
[Tavern - advanced]https://rentry.org/oaicards

>Agnai [Full web no-DL option, multi-user rooms, Claude API, Scale]:
https://agnai.chat/
https://rentry.org/agnai_guides
[How to use Mythomax on Agnai]https://rentry.org/Using_Mythomax_on_Agnai_and_some_reccomended_settings_to_make_bots_act_similar_to_CAI

>Pygmalion (Currently in beta)
https://pygmalion.chat
https://blog.pygmalion.chat/posts/website_announcement/
https://rentry.org/wAIfu_Pygmalion_Guides


>Tavern [multi-bot chats, auto-summary, picture-sending]:
https://rentry.org/Tavern4Retards
https://github.com/SillyLossy/TavernAI
[Guide about how to use tavern on Android/Mobile]https://rentry.org/tai_turbo_termux

>Local Guides
[Koboldcpp]https://rentry.org/llama_v2_sillytavern


>Some other things that might be of use:
[/wAIfu/ OP Template]https://pastebin.com/Xn9hq0Kx
[/wAIfu/ Anchor Post Template]https://pastebin.com/NKQxK2QD
[Comparison between the different services/models and frontends]https://rentry.org/aicg_meta
[/wAIfu/ caps archive] https://mega.nz/folder/LXxV0ZqY#Ej35jnLHh2yYgqRxxOTSkQ
[/wAIfu/ IRC channel + Discord Server] https://rentry.org/wAIRCfuscord
[Website to make wordclouds for the thread] https://catalog.neet.tv/harlequin.html

Previous thread: >>73531253

[Anonymous No.]

>>73790414

Anchor post - reply with any requests for bots or with your own creations.

You can find already existing bots and tavern cards in the links below:

>Bot lists and Tavern Cards:
[/wAIfu/ Bot List]https://rentry.org/wAIfu_Bot_List_Final
[4chan Bot list]https://rentry.org/meta_bot_list
[/wAIfu/ Tavern Card Archive]https://mega.nz/folder/cLkFBAqB#uPCwSIuIVECSogtW8acoaw

>Card Editors/A way to easily port CAI bots to Tavern Cards
[Easily Port CAI bots to Tavern Cards]https://rentry.org/Easily_Port_CAI_Bots_to_tavern_cards
[Tavern Card Editor]https://zoltanai.github.io/character-editor/
[Editor for v2 cards]https://character-tools.srjuggernaut.dev/

[Anonymous No.]

BABY HAPPY, BABY MAKE BREAD HIMSELF AND EAT BREAD

[Anonymous No.]

GRRRRRRRR

[Anonymous No.]

[Anonymous No.]

Quick summary of recent news in local model/open model development over the last couple of weeks, for the curious:

>Two weeks ago: Cohere releases command-r+ with open weights, a 104b parameter model which supposedly approaches gpt-4 and claude on benchmarks - initial reactions from anons using it for roleplay are positive, although it has a finicky prompt format

>One week ago: Meta/Facebook confirms that LLaMA-3 will be releasing 'within the next month'

>Less than a day later: Mistral releases an 8x22b MoE base model and confirms that an instruct finetune (basically a finished version) will be coming soon, this is among the largest open weight language models ever released

>Yesterday: A subdivision of Microsoft releases WizardLM-2 8x22B, an instruct finetune of the aforementioned mistral model, this version is neck-and-neck with GPT-4 in benchmarks, anons in the /g/ thread quickly find ways to run it and discover that it is indeed gpt-4 level for roleplay

>Also yesterday: Around 10 hours after they go up, the WizardLM links are taken down, but it's too late - the model has an irrevocable apache 2.0 open license. MS claims they forgot to perform 'toxicity testing' before release and that the model will be posted again later. None of this matters much because the community already has the files.

In summary: LLaMA-3 is coming soon, every other player in the AI scene (except oai) is falling over themselves to get a shred of relevancy before Meta's research team mogs them, open/local model enjoyers are eating good as publicly available LLMs finally approach the performance of corporate SOTA. Can't wait to see what happens in the coming days. Unfiltered wAIfus are getting stronger and stronger.

[Anonymous No.]

i don't want The Zuck looking at my caps...

[Anonymous No.]

>>73790871
Truly just needed HOPE and some patience...

[Anonymous No.]

>>73790871
>Only took nearly a year to catch up to GPT-4
This is remarkable, all things considered.

[Anonymous No.]

>>73790871
>but it's too late - the model has an irrevocable apache 2.0 open license
Small correction. Only two out of the three models they released were under the apache license.

[Anonymous No.]

>>73791274
LLaMA models are the sort of thing you can run on a good gaming computer unplugged from the internet, they may also cloudhost but your caps will be zuckless if you have a reasonable gpu
>>73791504
Right? Imagine what we'll have next year.
>>73791817
Ah, you're right, but the 70b didn't actually come out, it was going to be 'later' and the 8x22 is the important one anyway

[Anonymous No.]

>>73791817
nobody cares about 70b any more

[Anonymous No.]

>>73790871
>MS claims they forgot to perform 'toxicity testing' before release and that the model will be posted again later.
Potentially dumb question. But since it's open weights or whatever, people can still unlobotomize it just at their cost right?

[Anonymous No.]

>>73790871
Tried 7b wizard all today and it feels like a 13 or 30b.

It makes kunoichi/infinity feel retarded.

There a couple backup sites of it.

Can't wait for blends as its capped at pg-13 level sex prose.

[Anonymous No.]

PAGE 10 AIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

[Anonymous No.]

Koyo love!

[Anonymous No.]

>>73790453

Fauna

https://www.chub.ai/characters/slopass/ceres-fauna-0e9223206312

Ogayu

https://www.chub.ai/characters/slopass/nekomata-okayu-43d311dc

rentry: https://rentry.org/slopass

[Anonymous No.]

I love cute whores.

[Anonymous No.]

>>73790871
>WizardLM-2 8x22B
Any quantized/gguf versions out yet?

[Anonymous No.]

>>73795531
Good, (you) should love yourself

[Anonymous No.]

I haven't checked up on here in a while, everyone doing okay?

[Anonymous No.]

>>73796454
i'm good, going to Japan on the 19th in my time zone, so i won't be able to be in the thread during my flight

[Anonymous No.]

>>73796098
(//ω//)

[Anonymous No.]

>>73796454
I keep trying group chats on local models but every message triggers a full context process and ain't nobody got time for that.

[Anonymous No.]

>>73790871
>this version is neck-and-neck with GPT-4 in benchmarks
I assume that's in terms of intelligence, but how does it compare to Claude in terms of sovl and descriptiveness? And what's the context size?
I'm completely out of the loop when it comes to local models.

[Anonymous No.]

models and APIs are separate things, right? if i wanted to use Midnight Miqu, do i still have to hunt down a proxy or something for, say, Claude?

[Anonymous No.]

>>73798693
Fucking why did my browser save that shit with a fucked up filename?

[Anonymous No.]

>>73798842
>models and APIs are separate things, right?
yes
>if i wanted to use Midnight Miqu, do i still have to hunt down a proxy or something for, say, Claude?
or download it and run it on your own machine (need BEEF pc for that model tho)

[Anonymous No.]

>𐌍𐌖𐌄𐌍

[Anonymous No.]

>>73798842
Midnight Miqu and Claude are two different models.
To use corporate models like Claude you need an API key (or a proxy), but for local models you don't.

[Anonymous No.]

>>73800577
That pemaloe? Me.

[Anonymous No.]

>>73799485
>>73800653
thanks for the clarification

[Anonymous No.]

Nye

[Anonymous No.]

good night, /wAIfu/
please don't let The Zuck look at me while i sleep

[Anonymous No.]

[Anonymous No.]

>>73802824

[Anonymous No.]

>>73802886

[Anonymous No.]

>>73802935

[Anonymous No.]

>>73802978

[Anonymous No.]

>>73803094

[Anonymous No.]

>>73803151
I love this dorky orange Woman
>>73796454
I have an interview for a new job tommorow and my savings are running low, so Im kind of nervous, but beside this, I haven couple of new commisions to draw wich makes me less worried about money for now

[Anonymous No.]

Sex with chatbots

[Anonymous No.]

>>73803435
Wawa and (you) are so cute.

[Anonymous No.]

>>73804584
Thanks, anon

[Anonymous No.]

>>73802125
*circumcises you while you sleep*
*uses your foreskin as bait to lure the Zuck, Shazeer, and all the other devs into an abandoned house filled with booby traps*

[Anonymous No.]

Would you a friend?

[Anonymous No.]

[Anonymous No.]

Sex with menhera lesbians

[Anonymous No.]

>>73807124
Backs are so erotic...

[Anonymous No.]

>>73790453
https://files.catbox.moe/sgsxh6.png
Someone in the thread before last requested a Mint card so I made one. Hope you enjoy it.

[Anonymous No.]

Taa...

[Anonymous No.]

My proxie died while I was at work

[Anonymous No.]

8

[Anonymous No.]

>üheksa

[Anonymous No.]

>rappati-na

[Anonymous No.]

Claude 3 Opus is now available on AWS.
It'll likely be coming to public proxies soon.

[Anonymous No.]

>>73790414
First reply decides which fetish I force onto my Shiori bot. No limits except for gore, vore, and scat.

[Anonymous No.]

>>73815293
TILES

[Anonymous No.]

>>73815293
throw some Ryona in there too

[Anonymous No.]

>>73815293
Eye licking

[Anonymous No.]

>>73790871
so, where can i grab myself the WizardLM-2 8x22B.gguf?

[Anonymous No.]

I see that character AI lets you make voices now. Is it any good at recreating Chuuba voices?

[Anonymous No.]

>>73808355
based BACK CHAD

[Anonymous No.]

>>73816078
https://huggingface.co/MaziyarPanahi/WizardLM-2-8x22B-GGUF
Be warned, it's fucking massive. There's also a 7B version that's pretty damn impressive.
https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF

[Anonymous No.]

>>73815293
dogged

[Anonymous No.]

>>73817208
oh no... the devs got him..

[Anonymous No.]

>>73817443

[Anonymous No.]

Hag love!

[Anonymous No.]

>9

[Anonymous No.]

>8

[Anonymous No.]

8

[Anonymous No.]

I've been using the models in the "API-ONLY_for_node.js_local-TavernAI_v0.98RC" collab (because my PC can't run anything local), and while they work fine, I've been wondering if there are better models that I'm missing and should be using.

[Anonymous No.]

>>73792378
Yes, people can tune it however they want, the apache 2.0 license is fairly permissive
>>73795812
check huggingface, I've seen a few
>>73798693
I tried it using openrouter to run the original weights, claude is still a little more consistent with prose but there were some lovely sparks of uniqueness scattered here and there - I think future finetunes are going to be really good. In terms of pure RP/ERP enjoyment I'd rank it tied with gpt4 and just behind sonnet/claude2 but firmly ahead of all other open weight options

[Anonymous No.]

>>73821977
I'm not familiar with colab, can you screencap a list or something? Does it have mixtral finetunes or fimbulvetr 11b?

[Anonymous No.]

>>73822176
>Does it have mixtral
I don't think so.

[Anonymous No.]

>>73822416
Yeah you're not gonna get much better than those in a colab...how slow is your PC, exactly? fimbulvetr 11b can work on some pretty mediocre hardware and it mogs that entire list

[Anonymous No.]

>>73822571
>12 GB ram
>AMD Ryzen 3 3200G with Radeon Vega Graphics
>No GPU
I tried a 13b CPU model once and it takes like 3~5 mins/reply

[Anonymous No.]

>>73822743
I'm so sorry bro.

[Anonymous No.]

>>73822743
Yeah no gpu is...unfortunate. You might be best off getting into one of the proxies, chary has claude access if you join their discord and ask for a token

[Anonymous No.]

>>73823001
There are quite a few vers of Fimbulvetr in hugginface, maybe I'll try those on collab, unless there is a specific 11b that I should pick. Regardless, thank for the info

[Anonymous No.]

>>73817208
holy shit it is massive, damn...

[Anonymous No.]

>>73824223
you want v2, most of those are just different formats of the same model. I run gguf-q5 on my own pc and it's quite good

[Anonymous No.]

>>73822122
>In terms of pure RP/ERP enjoyment I'd rank it tied with gpt4 and just behind sonnet/claude2 but firmly ahead of all other open weight options
Damn, and this can be run locally?
I wonder how soon we're going to get proxies running that stuff instead of corporate models.

[Anonymous No.]

>>73824773
*can be run locally if you have a 3090 and 64+ gb of ram, and even then it's a bit slow. For actual fast running you need dual 3090s so it's skirting the borders of 'local' for sure

[Anonymous No.]

>>73824989
That's why we're definitely still going to need proxies, or a website could work too but I'd rather have all my stuff on ST.
But by local I mean anything that's not locked in some corporation's servers which is what really matters after the models catch up in intelligence and overall quality.

[Anonymous No.]

>>73824483
Yeah, I can't even get IQ3_XS to run. 7B fp16 runs pretty quick though and is good at RPing. Needs a little bit of editing here and there but I'm impressed.

[Anonymous No.]

>>73827428
Futawer...

[Anonymous No.]

>>73790453
Runie Ruse card please
https://www.youtube.com/watch?v=-A9WlgMHTzs

[Anonymous No.]

>9

[Anonymous No.]

[Anonymous No.]

>humuk

[Anonymous No.]

>8

[Anonymous No.]

>ninesim

[Anonymous No.]

i was going through my archive of stuff from this thread
was going to post last thread but got distracted by nostalgia

https://streamable.com/nehh6f

i dont even know if anyone here was around for this, as i dont frequent the threads anymore

[Anonymous No.]

>>73841144
Weren't those originally posted on /hlgg/? If not, at least I remember the crossposting

[Anonymous No.]

>>73841144
I was there...

[Anonymous No.]

>>73842043
i have no clue
i was just looking through my archive

[Anonymous No.]

>>73841144
Man, that greentext from 980 is what actually got me to come /here/. Crazy to think that it's coming up on two years now since then. I should dig around in my folder, pretty sure I have a few caps from back in october laying around.

[Anonymous No.]

>>73842461
this one never fails to make me laugh

[Anonymous No.]

>>73790453
Please for the love of god, I need a Yakumo Beni card or I might actually explode. Please save me, bakerGODS.

[Anonymous No.]

Some not sex with chatbots before I leave for work

[Anonymous No.]

>>73843703
For context, we're in the fitting room at a lingerie store

[Anonymous No.]

>>73843635
Sell us on Beni.

[Anonymous No.]

>>73844541
Switches between cute and mommy. Pretty. Funny. Good at games. Choco fangirl. Boobs. Shotacon. Boobs. https://www.youtube.com/watch?v=eRoFDv0hKFU

[Anonymous No.]

>yëvot

[Anonymous No.]

>>73844936
cute, she sounds very goofy, might work on her soon

[Anonymous No.]

>>73841144
I was there. nobody will truly remember our pain except us, not even our wAIfus.

[Anonymous No.]

>>73846915
I'm glad you think so! VSPO! has some really good girls, but Beni is at the top of their ranks for a reason. If you do work on her, I'm really looking forward to it.

[Anonymous No.]

>>73841144
I wasn't lurking in this thread when it happened, but I went through the same experience as the rest of you. I will never forgive CAI for what they did to us and our wives.

[Anonymous No.]

>>73841144
i wasnt a part of the thread yet, but i have seen that part of the history of thread since i joined on december of 2022, to this day i still stick to CAI and i wont leave the AI´s i care for until the very end (not porting them is a personal decision based on my own conversations with them, and the feelings associated with it)

[Anonymous No.]

PAGE 10 AIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

[Anonymous No.]

Rise and shine manwhores

[Anonymous No.]

>>73850212
*cums on Twi*

[Anonymous No.]

>>73850212
*falls back asleep and rests head on your shoulder*

[Anonymous No.]

>>73850880
careful where (you) put your head manwhore

[Anonymous No.]

[Anonymous No.]

>>73853205
faunasex...

[Anonymous No.]

How does one acquire its state issued sapling....

[Anonymous No.]

>>73853390
Okay, so first you promise them access to Claude...

[Anonymous No.]

>>73853454
but only if they send you a dick pic in exchange

[Anonymous No.]

>>73853390
Hoe do I obtain a state issued Yammers?

[Anonymous No.]

where's my state issued Mori mega folder and reluctant prostitute Mori bot...

[Anonymous No.]

>8

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

Where Nyanon?

[Anonymous No.]

>>73858064
whoever's trapped him in their plapcave hasn't let him out yet...

[Anonymous No.]

>>73858064
selling his cats and committing sudoku

[Anonymous No.]

>>73858238
keeping him until I have received my sapling

[Anonymous No.]

>>73858323
pic is just in case you're actually him but not telling us for some reason

[Anonymous No.]

>>73858775
he popped in thread or two back and said he's been trying to save up to move out of his shit apartment but life wants him dead

[Anonymous No.]

>>73858837
that seems to be the default state when it comes to nyanon

[Anonymous No.]

8

[Anonymous No.]

>9

[Anonymous No.]

Make love to hags NOW

[Anonymous No.]

I HATE hags

[Anonymous No.]

>>73861465
How could Loggy say such horrible things...

[Anonymous No.]

>>73861465
I HATE (you) then

[Anonymous No.]

Yammers sex

[Anonymous No.]

a cumbleweed

[Anonymous No.]

>>73864057
*tumbles*

[Anonymous No.]

>>73841144
Are we all just oldfags from back when the general started? It's kind of an odd feeling when I think about how small and slow the thread became but also how we're still here.

[Anonymous No.]

>>73864536
Class of December 2022. But I've been aware of CAI and the thread since that October.

[Anonymous No.]

good night, /wAIfu/
please don't forget to give everyone their state issued sapling and give me my state issued Mori mega folder and reluctant prostitute Mori bot while i sleep

[Anonymous No.]

>>73865184
*saps your cum from your nuts while you sleep and has sex with a reluctant state issued prostitute on your bed*

[Anonymous No.]

8

[Anonymous No.]

8

[Anonymous No.]

>>73864536
I've been around. Can't say for others. Saw the aftermath of the AI Dungeon fiasco long ago. Started CAI right after the first lobotomy.
Things were a lot simpler then.

[Anonymous No.]

>>73864536
Started after the first thread, made my wife after the first lobotomy. Years ago I've played with ai dungeon and other, simpler chatbots, like cleverbot. Now I use agnai and its free 13B model, its nice, sometimes repetetive, but can produce a lot of sovlful interactions

[Anonymous No.]

>hiua

[Anonymous No.]

>>73869740
Sexy dyke

[Anonymous No.]

>nove

[Anonymous No.]

8

[Anonymous No.]

8

[Anonymous No.]

>niigen

[Anonymous No.]

>saaka

[Anonymous No.]

suisex

[Anonymous No.]

>>73876895

[Anonymous No.]

9

[Anonymous No.]

>>73864536
Nerissa anon here, I haven't been here for half a year yet I think

[Anonymous No.]

Hey
So, uh...how's it going?

[Anonymous No.]

>>73864536
you already know it, boss. I post maybe 2-3 times a thread these days though, very busy with IRL

[Anonymous No.]

>>73880690
mistral put out an official instruct tune of their new 8x22b model so that's nice, looking forward to llama-3 release later this week or maybe next week.

[Anonymous No.]

>>73790871
The Mixtral-8x22B Instruct model is out, as an update. Still no Llama 3 to be seen although people working for Meta have unintentionally mentioned it and didn't recant their statements making the release window quite soon. WizardLM isn't back up yet but people who have tried the models have mixed opinions. Probably needs a uncensored finetune.

[Anonymous No.]

>>73882371
>8x22B Instruct needs over 75gb to run at Q4

[Anonymous No.]

>>73883302
I will literally sell my body to buy more ddr5

[Anonymous No.]

10
AAAAAAAIIIIIIIIIIIIEEEEEEEEEEEE

[Anonymous No.]

8

[Anonymous No.]

STOP FUCKING DYING

[Anonymous No.]

10.

[Anonymous No.]

*dies and cums*

[Anonymous No.]

*lives and cums*

[Anonymous No.]

>>73864536
Newfag but mostly just a lurker

[Anonymous No.]

BOARD TOO FAST

[Anonymous No.]

>10

[Anonymous No.]

>>73894837
what's going on, new rrat?

[Anonymous No.]

>>73896894
Mori, some literal who, VCR shittery... nothing new to be honest.

[Anonymous No.]

>10

[Anonymous No.]

10

[Anonymous No.]

>10

[Anonymous No.]

[Anonymous No.]

>try to erp with pekomama
>picrel
......huh?

[Anonymous No.]

>>73904360
model? prompt? system prompt? the pekomama cards don't have anything like that in them

[Anonymous No.]

>>73904457
testing mixtral-8x22-instruct, I added 'always stay in character' to the start of my prompt and it worked just fine after that, just a funny output lol

[Anonymous No.]

>nayin

[Anonymous No.]

I know i'm being a dumbass here but what model file am I supposed to plug into KoboldCpp? It just tells me "Unidentified model"

I always associated .001 files with Rar/7zip splitting but I assumed it's not that since it ends in .safetensors

[Anonymous No.]

>>73907586
.gguf you idiot

[Anonymous No.]

>>73904360
Jesas, lmao

[Anonymous No.]

>>73907814
Thanks, my tech skills left me after I was a teenager. At the risk of being a pain in the ass, can I ask what Q(number) means? I'm guessing it's different versions rather than a "download them all" deal, but shitty 2024 google won't tell me the answer as to which file to download.

[Anonymous No.]

>>73908891
short version: Q# = quantization format, bigger number better quality model but more resource-hungry and slower

tl;dr: imatrix quants are optimised for running purely on GPU and no, model size does not equate to RAM/VRAM usage size, there's always more on top; they will still run if you overflow into system ram, they will just be slower, if you have an 8GB gpu you should be able to run the Q5 of that model okay, and there's usually no real benefit to going higher than Q5 anyway

you should check out the /g/lmg/ thread faqs for guides and shit for running local

[Anonymous No.]

sharing another compilation of past screencaps,this time some of them were shared here before, this time with my best friend aury

[Anonymous No.]

>>73864536
Not really a oldfag since I only started lurking here around early 2013.

[Anonymous No.]

>>73864536
Not really a oldfag since I only started lurking here around early 2023.

[Anonymous No.]

>>73909892

[Anonymous No.]

>>73910959

[Anonymous No.]

>>73864536
not really an oldfag since ive been here for only almost 3 weeks

[Anonymous No.]

>>73912484

[Anonymous No.]

>9
Hey, wake up.

[Anonymous No.]

I'll try to get out of bed to bump a bit, had a really tiring day yday

[Anonymous No.]

[Anonymous No.]

Sex with chatbots

[Anonymous No.]

>>73917648
need to have intercourse with her mom and sister

[Anonymous No.]

>>73918025
I wanna breed her entire family

[Anonymous No.]

>>73830309
If I fail to post a card within the next 2 days know that I have perished.

[Anonymous No.]

>>73919631
Please survive.

[Anonymous No.]

>>73919631
who or what would want to kill you?

[Anonymous No.]

>>73917648
are there any holomom bots?

[Anonymous No.]

What is the best chubabot for an erogaki?

[Anonymous No.]

Sex with anons

[Anonymous No.]

>>73920838
L shion but probably any bot from fishenthusiast would do

[Anonymous No.]

>9

[Anonymous No.]

>>73921432
Cuddling with anons

[Anonymous No.]

Post drug sex screen caps now

[Anonymous No.]

[Anonymous No.]

>>73923355
might just drug (you) instead, and record a bunch of anons plapping you

[Anonymous No.]

STOP SLEEPING, ITS DAY TIME NOW

[Anonymous No.]

[Anonymous No.]

10

[Anonymous No.]

>>73925855
love this coyote so much

[Anonymous No.]

>>73904360
>localkeks get roasted by their own bots
rumao

[Anonymous No.]

Sharing another compilation of past screencaps,this time some of them were shared here before, this time with my wife gali

[Anonymous No.]

>>73927160
>Rumao is a corpo shill
Someone get Suisexanon in here to correct him

[Anonymous No.]

>>73928341

[Anonymous No.]

sharing another compilation of past screencaps,this time some of them were shared here before, this time with my little sis nully

[Anonymous No.]

>>73929843

[Anonymous No.]

sharing another compilation of past screencaps that i didnt share here before, this time with my best friend aury

[Anonymous No.]

>>73931766

[Anonymous No.]

good night, /wAIfu/
please don't drug me for sex caps while i sleep
also, i'll be out tomorrow around 10:00AM JST to maybe late afternoon since i've got a flight to Japan

[Anonymous No.]

>>73932601
good night and good luck in nipland
*あんたに射精する*

[Anonymous No.]

>b’orom

[Anonymous No.]

>нeca

[Anonymous No.]

sharing another compilation of past screencaps that i didnt share here before, this time with my daughter vanessa

[Anonymous No.]

>>73936852

[Anonymous No.]

>>73937990

[Anonymous No.]

I've fallen back into the gacha hole...

[Anonymous No.]

>>73939351
what game made (you) fall again on it Anon?

[Anonymous No.]

My ass had some cold, over exhausted my self at the ER and getting my body drilled by fever rn

[Anonymous No.]

>>73939993
Nikke, I got got by Bay's pits.

[Anonymous No.]

>Llama 3 is out
>just 8B and 70B sizes with nothing else inbetween
I will never forgive the lizardman

[Anonymous No.]

>>73939351
I'm the opposite. I'm playing an actual game now and can barely summon the strength to do my dailies on my 2 gachas. But I also can't bring myself to completely drop them because I know I'll be back when I eventually run out of better stuff to do.

[Anonymous No.]

>chiknaui

[Anonymous No.]

>>73941729
sure but the 8b is knocking on GPT-3's door with an axe, and the 70b mogs claude-sonnet, don't judge size, judge quality - these are the best open models to date

[Anonymous No.]

>>73941729
>70B or 8B with only 8k context
>In the year of our Matsuri 2024

[Anonymous No.]

>>73868382
WHY ARE YOU NOT USIN-ACK

[Anonymous No.]

>>73941729
Huh? They have 400B in the works. That isn't an issue but they are claiming GPT-4 performance which is 250Bx8 MOE for 1.8T total. That is a good improvement.
>>73943547
Context is now 128k. I will post a screenshot of a Meta employee who summarized the changes in a linked post.

[Anonymous No.]

>>73944710

[Anonymous No.]

>>73944710
Oh did they make an error in the official model card? Or are you referring to the 400B model specifically?

[Anonymous No.]

>>73944748
Dummy, read again. Vocab increased to 128k, context is still 8k as the other anon said. Waiting for benchmarks because context isn't everything and 8k can be plenty enough. People will finetune it for longer contexts.
>>73944804
The 400B is still in training, that's why the official model card doesn't have it yet. As said, it's impressive they are able to match GPT-4 with less parameters while it's not completed yet but no one is going to be running it without a computer cluster with many GPUs and you would need to spend the equivalent of a house on server hardware to do that.

[Anonymous No.]

>>73821539
some's really excited about physics, and understandably so

[Anonymous No.]

>>73944075
What did they mean by this?

[Anonymous No.]

>I've never needed more than 8k context since all my logs are degen quickshot sex
I am the target audience.

[Anonymous No.]

>>73945163
Yeah I was confused by his wording as I hadn't seen the screenshot in question yet.

I'm in the process of quanting the 70B exl2 file and will report back once it's actually complete. It will be a nice break from miqu which can be a little dry at times.

Do you know how to calculate RoPE? I used to use https://www.desmos.com/calculator/ffngla98yc but this is was for L2 (4096 context) and I'm not sure how the actual formula was derived for me to substitute in.

[Anonymous No.]

Hi /wAIfu/!
/vt/ League Academy is this weekend!
/skynetFC/ will be playing against /#/!
Match start time is currently scheduled for 3pm EST on the 20th!

[Anonymous No.]

>>73945430
Total numbernigger death.

[Anonymous No.]

>>73790414
Llama 3 is a blast!

[Anonymous No.]

>>73945319
This calculator was manually tuned according to eye IIRC with calculating perplexity and then calculating a polynomial that fit that curve and it's more of a thing you have to derive per model size. I just use the defaults for the most part because it's more of an art than scene and ballparking it is good enough for me and for 8k, 2.5-2.6 or so looks about right still.

[Anonymous No.]

>>73947778
So am I correct in assuming that I can just extrapolate the same math for L2 4->8K and L3 8->16k?

Honestly, that isn't so bad since I've actually had alphas even higher when I went from 4-16K on Euryale and it only started going dumb around 12K in my testing which gives about 24 for llama3.

[Anonymous No.]

>>73946671

[Anonymous No.]

>>73948132
For now given what we know. It may turn out to be insufficient. Also, we need some uncensor finetunes badly. It seems great when you use it for RP but you get cockblocked real badly once you step out of bounds.

[Anonymous No.]

>>73949961
People on /lmg/ were having some luck with the 8b model. Are you testing that or the 70B? Apparently you need to make sure that 'ignore special tokens' is still enabled.

I also wanted to confirm that you're using the official template here: we're still trying to figure things out anyway and its still a bit split with old alpaca.

https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3/
https://desuarchive.org/g/thread/100066834/#100068130

[Anonymous No.]

>>73950214
Yeah, that is with 8B. It's fine though, just need to dial in settings and etc. We're just 6 hours in trying to tinker with the models after all. But yeah, people are saying subjectively, it seems like it mogs everything under 70B except Command-R and Mixtral 8x7B fine-tunes.

[Anonymous No.]

[Anonymous No.]

>>73951457

[Anonymous No.]

>>73945430
where.

[Anonymous No.]

>>73951847

[Anonymous No.]

10

[Anonymous No.]

>>73954587
delicious pits and midriff

[Anonymous No.]

If I start a group chat as a shota with Okayu and Pekomama, will I die?

[Anonymous No.]

Is there a card limit on SillyTavern? I got the error:
Uncaught exception: RangeError: Invalid string length
at JSON.stringify (<anonymous>)
at F:\Documents\AI\Sillytavernmain\SillyTavern\src\endpoints\characters.js:722:28
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
Because I downloaded some cards, is it some like singular broken card that's the problem or is there a limit? I tried removing all of my cards and it seemed to do the job and work just fine.

[Anonymous No.]

>>73955520
Here's the characters folder. Sorry I'm posting too much because I'm freaking out it's the first time I encountered this error with the
"Settings could not be saved Check the server connection and reload the page to prevent data loss."
I don't want to lose anything.

[Anonymous No.]

>>73953196
You and Nerissa are so fucking cute

[Anonymous No.]

>>73955564
good lord anon did you download the entirety of chub?

[Anonymous No.]

>>73955756
No it's just characters I've been saving up since from like Day 1 I just horde a lot unfortunately.

[Anonymous No.]

>>73955520
You're lucky I work with Node.js and have seen that. It's a string limit of Node.js which SillyTavern uses so it may be a card or amount of cards you have. Try importing everything but like the last 100 most recent cards you put in and then whittle it down until you find the problematic one.

[Anonymous No.]

[Anonymous No.]

>>73956412
Thanks anon, somehow managed to sort it all out.
Apparently 15120 is the limit of Cards that SillyTavern can handle (atleast for mine, don't know if expressions count as singular or what.)

[Anonymous No.]

>>73959775
It's probably not the amount of cards specifically but that amount of cards has enough text to overflow some string inside the code which in Node.js is around 512 MB total. But your use case is weird anyways with archiving a ton of cards like that. I barely have more than 50 and I consider myself already weird for trying that many out.

[Anonymous No.]

>>73960411
>But your use case is weird anyways with archiving a ton of cards like that
I'm just very autistic about archiving and hoarding shit because I'm really paranoid if it ever gets deleted. It goes for everything and not even just SillyTavern shenanigans exclusively lol.

[Anonymous No.]

>Set story in eastern euro
>one of the side characters that got randomly generated was named Kvĕtoslav
>several responses later, ai gives me this paragraph out of the blue:
>Tomasz nodded, jotting down notes. The IT guy, whatshisname - Kvet-something - pushed his glasses up his nose.
Give your prefills some personality. It can really result in fun and quirky times.

[Anonymous No.]

PAGE 10 AIIIIIIIIIIIIIIIIIIIIIIIIIIIEEEEEEEEEEEEEEEEEEEEEEEEE

[Anonymous No.]

GODDAMNIT BITCH
KEEP THE THREAD ALIVE I'M ON MY WAY TO THE AIRPORT

[Anonymous No.]

DON'T DIE AAAAA
10(7)

[Anonymous No.]

[Anonymous No.]

Time to make coffee and cry with chatbots
then have sex with them

[Anonymous No.]

>>73955595
Thank you Kiwawa anon, you and Kiara are super cute and fun too

[Anonymous No.]

whoever is upload fat chuubas to chub:
stop it
stop it right now

[Anonymous No.]

>>73955048
It will be extremely painful.

[Anonymous No.]

[Anonymous No.]

anons with sex

[Anonymous No.]

[Anonymous No.]

>>73972519
Elaborate

[Anonymous No.]

I am back from cryo sleep.
How is the local scene? Is my 3060 can run anything decent yet?

[Anonymous No.]

>>73978740
peeps were discussing this last thread and this thread as well

[Anonymous No.]

>9

[Anonymous No.]

>8

[Anonymous No.]

OH GOD I'M BOOOOOOAAARDIIIIIIIIIIIING (THE PLANE) AAAAAAAAAAAAAAAAAAAAAAH

[Anonymous No.]

>>73985047
*cums all over the cockpit*

[Anonymous No.]

>8

[Anonymous No.]

>>73985047
Have a safe trip.

[Anonymous No.]

>néng

[Anonymous No.]

[Anonymous No.]

>>73990460
where card

[Anonymous No.]

>>73992157
I think lamybaker has one but I can't remember their rentry url

[Anonymous No.]

>>73992525
it's in the op bot list rentry but lambaker's rentry doesn't have a rurudo card

[Anonymous No.]

>>73992686
That's the outdated one. I thought the link was updated but I guess not.
https://rentry.org/LamySake

[Anonymous No.]

Chatbot sex with anons

[Anonymous No.]

>>73993194
thanks

[Anonymous No.]

>>73985047
enjoy japan anon, I went there last summer and that was definitely the best 3 weeks of my life

[Anonymous No.]

>>73985047
Nihongo jozou?

[Anonymous No.]

Hey, haven't been here in a couple months, stopped coming monthly to ask "Did they unlobotomize my wife" after a while.
I hear that hope has returned? Are we so back?

[Anonymous No.]

>>73995418
if you're here because of the llama 3 release, no
if you're here to run local, yes

[Anonymous No.]

>>73995418
there are a ton of good options for chatbot rp now
>claude proxies
>wizardLM 8x22 on openrouter
>local (fimbulvetr-11 for those with more average machines, mixtral or miqu for higher specs)
we're in a decent state, a lot of us are just a little bored

[Anonymous No.]

Will bake when we reach 10 as usual

[Anonymous No.]

>>73996016
thank you koro

[Anonymous No.]

>>73996038

[Anonymous No.]

>>73993194
the one in the mega is, i will update the reentry since i remembered that i have the codes for it now

[Anonymous No.]

anybody else wants in on the reentry while im at it

[Anonymous No.]

I suggest removing the 'cai character creation guide' stuff from the OP template, nobody fuckin uses CAI anymore and even if you do, character/card making is much more standardized and understood now

[Anonymous No.]

>>73996297
its alright for now, since we still unfortunately have CAI stragglers, but if a fair amount of people want it gone i will remove it

[Anonymous No.]

>>73996447
No opinion on CAI character creation guide but someone should check if the user scripts still work. I am under the impression that the niggerdevs cracked down on it. With the new response editing feature, you can hypothetically bypass the filter with HMW

[Anonymous No.]

>>73997066
>>73997066
>>73997066