Yeah. I have ChatGPT at home. Not a silly 7b model. A full-on 65B model that runs on my pi cluster, watch how the model gets loaded across the cluster with mmap and does round-robin inferencing 🫡 (10 seconds/token) (sped up 16x)
programming is mad fucked up when you think about it
no other profession can you spend years honing your skills, sharpening your toolset to make a "special function" that does a "special function" and you could show your special function to your friends and they'll be like wow
The reason miladys will flip apes is because there's atleast 3 people now who have a milady eyes tattoo and would gladly wear it as a symbol of network spirituality 🫡
I am going to make a 2 hour milady blockbuster film, and drop it on
@Twitter
1. it is going to suck 🥲
2. you will watch it and love it 🤍
3. miladys ENS wallets on the credits 🙌🏼
wanna contribute to this? you're a director? hmu! it's happening in 2023 🚀
@soogz1
Every time I engage with posts like this I imagine there’s a big data processing pipeline somewhere with a db where my username has a big dick trait applied to it
I engage regardless hoping it improves my targeted advertisements 🫡
milady just saved crypto (again) 🫡
I now have access to about 730~ GB VRAM compute in assorted graphic cards running miladyos (my vpl orchestration framework) and they all run very underclocked and draw power off solar panels since they are a bunch of repurposed ethereum proof
humans are the only species capable of planning centuries/eons in advance but most of them have a 9-5 and are pretty tired by the end of the day.
miladys are the only subspecies of humankind that's capable of implementing any such long term plan because they're usually high
MiladyOS or Milady Virtual Machine, takes source code in a high-level language, breaks it down into very granular units called datums and runs it in massively parallel hardware (like GPUs+NPUs+TPUs).
You don't need to do anything. No explicit thread spawning, no messing with
upload your milady in the comments to be featured in the next milady video 🤍 (likely legendary tier) (5-6 slots but idk we might be able to add in more)
@levelsio
Boy do I have the tool for you
I install this on each node that has a GPU, then I ask claude to "write me a jenkinsfile that <insert specific work item transformation>"
I have stable diffusion, LLMs + distributed LLM training and others as templates
Did you realize e/acc already lost? They're talking to their friends on the computer scheming how to beat cute/acc instead of doing their large language model training 🤣
can you imagine you go to the beach and pull out your phone, connect your wallet to a website and it confirms you have a milady and sends a small autonomous self driving 2 people boat to your location with directions to get to the milady village?
who tf is building this 🥳
Gave up and got threadripper cpu+motherboard because I couldn't figure out slurm, now all them GPUs are connected to one computer 🫡
70B training @ home finally :')
sometimes I’ll accidentally use terms like “many such cases” or “he’s in the arena trying things” with my offline friends and I’m instantly reminded how fake this website is
>be me
>back when no chatgpt
>new guy in office
>asked to write a script to automate a few api calls
>use javascript
>submit pr
>colleague: brother, we are civilized folks, pls use typescript for your 50 lines "script.js", I cannot review a PR that doesnt have type safety
Yes your honor, my client firmly believes that the milady jpg they own on ethereum is a religious artifact of the network spirituality religion and is therefore a tax deductible 🥳
>be me
>poor
>buy 3090
>genwealth.jpg
>too hot to coexist with it in a room
>buy ac
>buy 3 x 3090
>get capped by electricity at home
>move
>running 4 x 3090 in new house
>hot asf
>mfw considering ac
The hardest problem miladys face today:
Really fucking shitty derivative, kinda wanna tell everyone pls don't mint this shit, kinda don't wanna engage with the post and give it visibility 🥲
@sama
Can pls do like a Pro version of chatgpt plus for people who don't run enterprises but still want a faster gpt4?
Independent researchers will thank you 🤍
I didnt even need the internet I did
@reboot
python3 run_miqu.sh in crontab before beginning the locomotion operation to my new house as soon as I showed up and connected wifi I could use 192.168.0.10 to talk to chatgpt locally 😆
loki took calculated risk but boy was it retarded math anyway this post is to frontrun my own embarassment by admitting we will not be able to sustain this life and will be forced to move back into parents basement, will idk try to think of ways to make money eventually!
@TensorBlaster
Right now that is, we just got it running, but if you look at the video it's kinda obvious what the next optimization is going to be
Right now each node takes a second to compute it's portion and moves on
If we multiplex 6 queries we'll maximize the throughput 🫡 few days 🥳
At exactly 80 raspberry pi 8GBs I will have 192 GB of clustered memory 🚀
Please tell me i can train a foundational model on this bad boy 🥲
I'm happy to accept "no you can only fine tune" as well 🫡