№17170671[Quote]
What if you make a bot like that 1 guy did on 4cuck
№17170681[Quote]
>>17170671idk how to do that doe
№17170708[Quote]
HOLY GLOW
№17170742[Quote]
>>17170628 (OP)There's a dataset of the entire ki on hugging FAC, op just finetune like qwen 3.5-3b with ur set and that dataset
№17170780[Quote]
>>17170742i will release a 4b variant but I dont think i can use the hugging face one
<whyits not structured correctly, you need it to be in q&a format, I ran my dataset thru an ai tool so it only keeps those that actually make sense (eg, reppys that are just bvmp, or are not related at all get thrown out) its more important to have a small yet high quality database
also I use google colab to train it (they do it for free)
№17170811[Quote]
where to use
№17170812[Quote]
>>17170742also i tried cleaning it up since the raw text has things like {{this}} and [this|o algx] but i wasnt successful and im pretty sure I couldnt fit everything in my 4hr training window der colab gives for free
№17170815[Quote]
Seca
№17170837[Quote]
Who could take more bbc up the ass spadeson or femquote
№17170839[Quote]
>>17170811model info in /tech/ tried making a website for yall but i was having issues and i didnt want to accidently dox myslef
№17170876[Quote]
>>17170839i got a virus dont download
№17170880[Quote]
What south asian country is wonned?
№17170921[Quote]
>>17170897who are pantied and sissified by bharatBVLLs?
№17170935[Quote]
BRO I ASKED IT A QUESTION AND IT SAID MY FUCKING ZIP CODE
№17170943[Quote]
>>17170935its anonymous bro they are hacking you
№17170961[Quote]
>>17170921>The Indian people, or at least the ones who have heard of the site and got inspired to make a new oneits incomprehensible….
№17170967[Quote]
what do you goon to the most?
№17170985[Quote]
>>17170780Use unsloth, they have a google colab template and its meant for qa dbs
>include the Ip leak №17171014[Quote]
are you a black tradcath?
№17171068[Quote]
What is the gem that saved the party?
№17171078[Quote]
>>17171068>The Nucleon crystal, which can save the party if it is used correctlymarge
№17171120[Quote]
>>17171020These are just their dynamic quants, I'm talking about their training thing, a friend of mine has a few servers that have like 9 rtx pro Blackwell 6000s, I'll probably train qwen3.8-35b-a3b with h both datasets (using a low kld uncensored / a liberated base)
№17171142[Quote]
>>17171120im pretty sure i just use an unsloth model with sft training on the dataset idk i just found out how to finetune 4 days ago
№17171164[Quote]
>>17171120For local use,
Maybe the biggest he can train on his servers is dsv4flash? Takes a lot more resources to train compared to inference
№17171165[Quote]
Ask it which os the most goonable
Spadeson
Femquote
Femjanny
Cobblestone(the former ki administrator)
Soytan
№17171166[Quote]
>>17171120imagine there was a prismml 35b a3b
>qwen3.8 35b a3bmarge….
№17171209[Quote]
i dont understand the techbabble in this thread are you guys trying to develop a sentient nusoi ai or something
№17171220[Quote]
>>17171164But ehh he will be running glm 5.3 (literally trades blows with mythos in cybersecurity.. How?)
>training glm 5.3 on all of these datasets and making the ultimate soyteen oxbot №17171229[Quote]
>>17171166Acciideentially confirmed it
№17171238[Quote]
>>17171220nigga i have 2gb of vram
№17171247[Quote]
>>17171209training one that acts the most like one
№17171258[Quote]
>>17171166Bonsai is kinda shirt tbh
What was it called?
Maple-preview or mach-1
Nvm it's mach-1
Mach-1 is literally what you are asking 4
№17171270[Quote]
Use the almost 300 thousand post io database leak that some random tranny leaked to train it
№17171271[Quote]
>>17170628 (OP)use it as a babybot
№17171278[Quote]
>>17171238Use lfm2.5-2.6 q6
№17171291[Quote]
Use gpt2
№17171293[Quote]
File: t.webm 📥︎ (3.49 MB, 416x734) 75484e93c7f407a3d5da26dc85279a0f0ImgOps
№17171316[Quote]
>>17171278idk whats the end goal here doey, it doesnt take alot of params to build a good roleplay model. as for doxxing doey? idk for that you would need to reverse engineer the api call for websites like whitepages or something
№17171343[Quote]
ask it how many minutes nate would last if he was sodomizing femquote
№17171346[Quote]
>>17171316Just give it a harness, write some skills, give it a docker container with all of your osint tools installed + how to query breach.VIP / cloudsing for intelx search (paid)
№17171468[Quote]
>>17171300heckin wholesome captain capeshit montage of xim protecting jew york (or just any other FOTMville ad niggsemen or however it goes) and teaching the fuggen bigophobes liberal values because good goys ALWAYS win YEASSSSSSSSSSSSSSSSSSSSSSS
№17171473[Quote]
>>17171369I've done this, also found out how to get the clean text from a ki article.
It first checks if the ki article exists in the scrape folder or if the scraped article is a week old
If so it fetches the page, cleans out the garbage etc
Also if it sees text like [[this]] or [[nigger|quote]] it knows its a reference to another article and it fetches that
It does rag with these articles, also in its memory system (hindsight rn) I've got it to memorise a few terms
№17171578[Quote]
>>17171473did you use the wiki as a dataset or a tool? mane my brain bouta explode n sheit
№17171613[Quote]
>>17171578Not a dataset since im not training it, just saving as .txt and the model reads the page
№17171841[Quote]
>>17170681Yea cuz your brap brained