Bill

@LottoLabs @PopVerseYT @0xSero @sudoingX DM me if you need anything happy to help.

@PopVerseYT @BillQueens I think local llama on Reddit and @0xSero or @sudoingX have a x community. If you got a 5090 get 27b running on there and install Hermes agent. (It’s not sota but it’s pretty good)

Hermes Agent + qwen 27b This project is to see if I can get a fully functioning and safe saas/business set up w/ hermes and 27b just off telegram. Site is pretty much done, auth is done, just hooking up stripe payments and manually auditing. Will probably do a sota audit with a couple different models to see if I miss anything.

@sudoingX My wife tries this on me everyday

Reinforcement learning

@LottoLabs Ya I was so sick of vLLM horseshit lol. I just built my own frontend today to control everything.

@BillQueens Fair I’m going into those weeds now 😭

@LottoLabs I’ve just been ripping llamma.cpp so GGUFs have been solid for me. Had so many issues with vLLM nightlies said fuck it and haven’t looked back.

@BillQueens I think I ran that quant on a rtx6000 pro and it obviously ripped, I like the 27b overall though, mainly used the unsloth, might as well try nvfp if you have Blackwell chips

@LottoLabs GGUF will look at trying this tomorrow - what are your thoughts so far?

@BillQueens Have you been running the NVFP model? huggingface.co/Kbenkhaled/Qwe…

@hyperprior @0xSero donate.sybilsolutions.ai

@0xSero where’s the link to donate?

Man, what the hell. That's a few years salary for most of the world donated in under 24 hours. I promise I will do everything in my power to make this worth it for all of you.

@sudoingX 32GB

how much VRAM do you have right now

@BoomersBetz I bet Mormans -2.5

Alright ive cooled down. Whats the lock

@ivanfioravanti It’s a dual license ya dingus

I see in uv repo on github there are both MIT and Apache 2.0 licenses, which is the right one? 🤔 Relevant if someone wants to create an OpenUV fork...

@leerob what’s the one more thing ?

Welcome to selling insurance TCU

@Teknium My dev box is just a server basically. Threadripper and 5090 - 128GB ECC with Ubuntu on it.

Do you agree? Where do you do most of your dev work? I personally have a linux server machine in my house that I run hermes-agent on and do my dev work from. How bout you?

What an upset! My bracket is in shambles

@TeksEdge In what world is any of this accurate

@selfhostedmind @maria_rcks Likely the quickest best option available in her country.

@maria_rcks did theo really cheap out on a refurbished 2022 M2 air ?? wow so generous

I can't believe im typing this on a macbook, everything feels really smooth and weird (macos) but really really thank you so much theo, this is amazing, you really didnt have to, but thank you so much ❤️

@HedgehogZone_ Mine passed away

@DailyHitman I do

Spent two days straight staring at my wall visualizing the tourney. Who wants the first POD? 🧘‍♂️

@__tinygrad__ Imagine not building shipping container token machines in the year of our lord

People are too focused on trying to build God and not thinking about the unit economics 🤑

@JaMarc0 @andrewdfeldman @cerebras They don’t have the supply chain.

@BillQueens @andrewdfeldman @cerebras but thats just a financing issue

NVIDIA's biggest GTC announcement was a $20 billion bet on the same problem we solved 6 years ago. Their next-gen inference chip - not available yet - has 140x less memory bandwidth than @cerebras. To run a single 2 trillion parameter model, you need 2,000+ Groq chips. On Cerebras, that's just over 20 wafers. Even paired with GPUs, Groq maxes out at ~1,000 tokens per second. We run at thousands of tokens per second today. And every day. In production now. Why? When you connect 2,000 chips together, every interconnect has latency. Every cable has overhead. It doesn't matter what your memory bandwidth is on paper if you're bottlenecked by the wiring between thousands of tiny chips. We solved this with wafer scale. One integrated system. Little interconnect tax. Jensen told the world that fast inference is where the value is. He’s right - it’s why the world’s leading AI companies and hyperscalers are choosing Cerebras.