Alexandre Strube

HPC, Supercomputers, AI research. Random stuff and blurry pigeons

[Jen Zhu @jenzhuscott]

Legend…. ❤️‍🔥

[International Cyber Digest @IntCyberDigest]

❗️🇺🇦🇷🇺 Ukraine just open-sourced Russia’s entire arsenal.

[Alexandre Strube @alexandre_ganso]

@notrealmuch @alexocheema I have been serving thousands of users with VLLM. Do you think I have much to gain by using nvidia's stack?

[Sebastian Caliri @SebastianCaliri]

I just tested my hand in a mini version of this scanner. Images that are higher quality than MRI, whole body captured in <1 minute, virtually free to run. This is going to change medicine.

Things get even crazier when you consider the possibility of using the same tank to focus ultrasound to ablate tissue, stimulate nerves, etc.

The FDA is not in the slightest ready for this. People will also complain about incidental findings but they are wrong and don’t understand how quickly software can improve and how inexpensive a time series of scans will be to generate.

[Commodore Computer Museum 🕹 @MuseumCommodore]

How many people did it take to make the legendary 1991 cinematic masterpiece Another World?

Basically, 1 person!

Another World was made almost entirely by one single developer — Éric Chahi — on a stock Amiga 500.

He coded the entire game in assembly, created his own polygon engine, drew all the art, and even built custom tools. Only the music was done by someone else. (Jean-François Freitas... so I am guessing 2 people all up LOL!)

No dialogue. No text. Just pure visual storytelling with fluid rotoscoped polygon characters that still look incredible today.

Who else was/is a fan of Another World?

[Alexandre Strube @alexandre_ganso]

@dannolan Turns out they didn’t. And to make things worse, you can’t back up an iPhone with an unresponsive screen - because it asks for the number code. So you can’t even make a recent backup before replacing it.

[Lisan al Gaib @scaling01]

very good

they removed all the saturated slop benchmarks that were inflating chinese models

[Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTex]

A really massive release from Qwen. This is the Nvidia/GDM turf.

[Alexandre Strube @alexandre_ganso]

@TheAhmadOsman I need 128 a100 40gb to run it in full context and every single user of my platform who tried it said it was crap. I am using the gpus for minimax.

[Alexandre Strube @alexandre_ganso]

@elder_plinius Are they going to ban my gpus outside the us?

[Alexandre Strube @alexandre_ganso]

We peaked at the Harlem Shake. It's been downhill from there

[Alexandre Strube @alexandre_ganso]

@LivingTricks_ I like those progression plots, when they are grounded in reality. Pay a better version of LLM to fetch data.... or just do it yourself. This is stupid.

[Alexandre Strube @alexandre_ganso]

@catalinmpit ragebait much?

[Alexandre Strube @alexandre_ganso]

@MiniMax_AI @NousResearch https://t.co/uNZGv2UjbS

[Alexandre Strube @alexandre_ganso]

If you are using @MiniMax_AI 's M3 model with @NousResearch 's Hermes, you will see the reasoning process happening, as Minimax now has a new tag `&lt;/mm:think&gt;`. You might apply my PR to your Hermes so it works properly: https://t.co/ueaY0NgXHo

[Alexandre Strube @alexandre_ganso]

@aditiitwt To me right now is minimax m3. Kimi is too big and too slow. I have it running on 256 A100 40gb gpus and it's dreadful.

[Alexandre Strube @alexandre_ganso]

@jerryharlow @aditiitwt No

[Alexandre Strube @alexandre_ganso]

@spandrell4 Because those get outdated REALLY fast. Have you been following how fast LLMs have been changing?

The model he runs can run in pretty much anything, it's tiny - as this is a proof of concept.

Try a real model and things get expensive really fast

[Alexandre Strube @alexandre_ganso]

@DavidOndrej1 @0xSero Because they are young enough to to equate nvme with storage.

Plus, minimax takes one hour to load from a parallel Fs with hard drives

[Alexandre Strube @alexandre_ganso]

@himanshustwts Because it isn't - they are running a plugin of SmiReasoning

[Alexandre Strube @alexandre_ganso]

@bbbrezenski @Prefeitura_Rio velho eles aplicaram um plugin