正在加载视频...
视频加载失败
Demo for the sharded inference engine on: gpt-oss-120b on 4090's, max input context 95k, max output context infinite (tok/s degrades with context) queue of max 12 prompts concurrently, use it wisely
91,357 次观看 • 4 个月前 •via X (Twitter)
37 条评论

for those who doubt: IP's of the nodes are verifiable code for this demo is found in the github:

Running a 120B model split across 4090s over WAN with verifiable receipts and no central server is genuinely unhinged infrastructure. This is what decentralized compute was supposed to mean before everyone got distracted making JPEG coins.

@c0mputeAI bro’s hard at work

@mehdixxxz @c0mputeAI can't change the DeAI space by doing nothing all day

Insane 👏

this is so fucking cool

we're pioneers mohit

Homie grinding hard 24/7

@toly @mert @solana where are you??

a ture pioneer, this is sick @leyten doesnt stop building

@toly @mert @OpenAI Check whats building here

infinite output context is doing a lot of heavy lifting in that sentence. 'tok/s degrades with context' is just 'technically infinite if you have three weeks.'

this is very cool i wish you would post on the main account, low visibility here :( very cool site and i liked how it answered what i asked wp dev

Bro casually making the next frontier of DeAI a reality

few questions why only 4 workers are participating? website says 300+ and who is the user here? website says 0 busy is this a separate experimental demo from what is live currently?

Good work sir, this could be a real solution to the crazy high server costs in AI inference

@shawmakesmagic u might like this one

Can you add a chat history rq. shit is cooking bangers

Probably nothing...

Cope and seethe haters! Builders just keep building until there’s no more fud left 🫡

people aren’t yet ready or can imagine

This tech is insane! Major props to the dev @leyten @solana @toly

very easy to see

⛓️💥

这是证明自己?

by the nodes scattered across the USA

@c0mputeAI @pmarca @elonmusk @TheRundownAI @rowancheung @mreflow @karpathy @ylecun @sama @emollick @demishassabis @toly @solana @blknoiz06 New AI tech and dev on the solana:So11111111111111111111111111111111111111112 scene the top dogs need to know asap $zero @c0mputeAI

nice, keep it up. Do you have a timeline for progression of full scale utility from this demo. what will fully working setup look like

amazing work

👏

Respect😍

12 concurrent is tight. What's latency under full queue for realtime voice?

@c0mputeAI @toly @NVIDIAAI @nvidia

W

🔥🔥

🌶️🌶️

remarkable work. Can’t wait till the day when people can use your tech to form p2p network and both providers and consumers can benefit from it. Right now it’s still not cost-effective enough though but I truly admire this work!


