all a / b / c / f / h / j / jp / l / o / q / s / sw / lounge cgi up wiki
[Home] [Catalog] [Search] [Inbox] [Write PM] [Admin]
[Return]

Posting mode: Reply
(for deletion)
  • Allowed file types are: gif, jpg, jpeg, png, bmp, swf, webm, mp4
  • Maximum file size allowed is 50000 KB.
  • Images greater than 200 * 200 pixels will be thumbnailed.
File: Interesting_image_to_post_225.jpg
(238 KB, 1920x1080)[ImgOps]
243712
No.158415
▶
hello
i would like to create my own AI chatbot from scratch and train it on heyuri posts, but i dont know where to even start
the most that i get by searching online is people telling me to use existing models to do this and that, but i want to make everything from scratch (everything!!!) and most importantly I want it to run locally on my computer, the problem is that i dont know anything about the topic
if you have knowledge or resources that I can look up that could help me please share them here even if it's just to tell me i'm a fool for even thinking of doing something like this and that i should gently place my cheek on the rails and yell "here comes the train!" mere seconds before my death at the young age of 45

please observe the interesting image i have uploaded to entice you to reply to the thread
Marked for deletion (Old)
>>
No.158416
▶
have you tried asking an AI chatbot?
>>
No.158417
▶
>>158416
thank you for the swift reply
yes i have tried asking an AI chatbot but it kept talking about difficult topics that I do not understand, what i understood was some points in what i'd have to do to make an AI chatbot (for instance turning words or sentences into "tokens" which are mathematical representations of sentences so that the program can understand them) but upon asking for more details the chatbot would keep giving even more difficult concepts which required even more questions to explain and so on and so forth, so I figured it would be best if I could find some sort of archive that contains all the knowledge I need and explains all the topics instead of asking a chatbot questions over and over again like that
>>
No.158432
▶
>i want to make everything from scratch (everything!!!)
that'll cost at least a few million dollars and if you have that kind of money you can hire experts to do the complicated technical parts for you
>>
No.158457
▶
>>158432
hello
could you explain why it would cost a lot of money? i was under the impression that i could program the AI myself like that one time i made a program that could estimate prices using gradient descent
>>
No.158468
▶
>>158457
I don't know enough about it to write a good explanation.
>and most importantly I want it to run locally on my computer
How much RAM and VRAM do you have?
>>
No.158539
▶
>>158468
i currently have 24gb of ram but i'm willing to expand it, i "only" have 8gb of vram but i cant upgrade it because buying a 2k gpu is not an option for me
>>
No.158542
▶
>>158539
>8gb of vram
That's barely enough to run a heavily quantized (retarded) LLM locally, much less to train one. It's enough to generate very high quality images with NoobAI/Illustrious models and even train loras for them, so maybe you should consider training a lora to generate heyuri-tan pics or something instead of a chatbot. I don't know how much resources are required to train the LLM equivalent of a lora, so that might be an option if you don't mind the chatbot being a little retarded.
>>
No.158543
▶
>>158542
>if you don't mind the chatbot being a little retarded
can you quantify "a little"?
>>
No.158544
▶
>>158543
I've never used a chatbot AI so no. I know quantizing AI models makes them significantly dumber. I've seen comparisons of images generated by Qwen Image Edit's ~40GB model and its ~8GB retarded version and as expected they look quite a bit worse. A retarded chatbot trained on heyuri posts could be funny though.
>>
No.158545
▶
>>158542
>That's barely enough to run a heavily quantized (retarded) LLM locally
just enough to run a Heyuri LLM! :biggrin:
>>
No.158629
▶
Admittedly I don't know much about generative technology, but is there a specific reason why training and running models usually involves GPUs and VRAM? I would think it would be possible to run it off CPUs and regular RAM, even if performance and speed suffers.

Also any recommendations for open-source FOSS implementations of chatbots, image generators, video generators, voice generators (I don't know the effectiveness of combining it with OpenUtau) etc.?

Your fortune: Only time will tell

>>
No.158643
▶
File: 1668902507591530.jpg
(1.1 MB, 3014x2686)[ImgOps]
1150976
I am back, everyone
After pondering the question a bit i decided that i would accept using already available tools to achieve my objective, the requirement that the end product should run completely locally still applies, however, so I ask what should I be using? I would like to do as much work as possible with my own tools on my own to increase my knowledge, but it's ok if that is unreasonable
To reiterate I am looking to make an AI chatbot that resembles a typical heyurizen as much as possible, I am currently not interested in generating images as I find AI generated images to be poor quality (usually)
>>
No.158644
▶
thanks for the neko picture
as for your question, I'll leave it to someone more knowledgeable... there was a guy here who used his local AI to share us a loli pee lemonade recipe a while ago
>>
No.158647
▶
how to have local ai..
>>
No.158662
▶
>>158647
You can use Ollama. if you're looking for local models i think ollamas own repos might suffice
Delete post: []