#programming
1 messages · Page 508 of 1
conversion from?
1 billion Hz to GHz?
im asking why you claim it makes a sound at that frequency
I just pulled a number from my hat for a joke
Doesn't need to be realistic
you can actually calculate it, assuming the size of an air molecule is about 3*(10**-10), and the speed of sound is 343m/s, the theoretical limit on frequency is something like
343/(3*(10**-10)) which is about this number
superboxs brain plays a H#₈₀₀ constantly
proof: nobody
which is about 1 THz
above that, the wavelength of the sound is shorter than the molecular width of the air and the wave becomes incoherent
hi
at which point its essentially heat rather than sound
hello
hi
yall r smart
however, in practice, you also have to consider attenuation
hello!
false actually but thanks 
pretty sure everyone in here can reliably say they are dumb at something
im assuming that while 1.14 THz still works, it doesnt travel very far
Terahertz 
even at 10mhz, the attenuation could be upwards of 1000 dB/m
does anyone know abtt the golden record?
yes, terahertz
how many GS of acceleration would you have on a 1ghz speaker
you'd need an earth-shatteringly loud sound
depends on how loud it is
what yall discussing abt?
still quieter than the scream babies make on planes ngl
how distant do you want to be able to hear it from
no
sound
that stuff is rlly cool for me
k

1cm works for headphones i guess
huh
just need slightly better hearing range
at 1GHz
yall cmp. sci. students?
so the number is like
im a baker
im gamedev, but i do a bit of other stuff too. mostly what interests me
I have a job :oldge:

i think my brain would explode

ako
"audible"


not just your brain
what yall talking bout sound?

the entire universe would collapse instantly
at that point its a human ear skill issue. jsut evolve better ears
and become stronger than the universe

Yes I’m familiar with the golden data storage record attached to the voyager 1 and voyager 2 spacecraft before their 1977 launches
time is subjective we should try doing this
What of it?
i love all that stuff
carl sagan
some ppl say it's a bad idea.
to be clear, the gravity of a neutron star is about 10**11 g
how many watts would it take to drive such a speaker 
but the vastness of the universe makes it impossible nearly
more energy than exists

I have energy+1 so we are good
basic principles of very very big number
how beeg
what yall talking abt sound?
i cant spell for shit
high frequencies well above the range of hearing

That happens at 238 db if I’m not mistaken
no
damn
i refuse
I'm feeling lucky, allow broken
if youre accelerating 1 gram over 1 metre you'd need about 10**9997 joules of energy
does anyone know how to find diff. servers
this is not googol dot com
ye there is
"Discover" button at end of server list
tysm
only need
10**9989 neutron stars in one place

add one for good measure
that was a new discord user
missed a zero
its weird to realise people stumble into there as their first discord experience
thats more than we have available ye
a little
imagine the bandwidth though 
that doesnt account for how variable universe's energy levels can be tho
we can make it work, im pretty sure some spiderman thing had multiverse energy theft
zero
mass-energy is positive and
if it variates outside of the range of 10**70, please be my guest to show how that makes a dent in
10**10000
no all of it should be for gpus and ai /s
wrrr
Spider-Man: Across the Spider-Verse (2023).
if spiderman can do it so can we

copyrighted material boo
im pretty sure multiverse theft isnt outlawed anywhere
how can the sacrifice of universe incrrease wifi speed?
Damnn
Batman has a 20000 decibel radio so you can use that
can anyone plz explain wth is going on
194db not 238db
wrr
this is the programming channel, here we're doing nerd shit
like what sorta thingy
as I said llama.cpp would also work if you really want native without containers
it's also in Nixpkgs
shadow says lamas are bad
I think we should make microwave oven but use QHZ (quintillion hertz) instead
Shadow says everything is bad
That’s why shadow doesn’t have a clone of himself yet
And why you will

just use llamacpp brother it’s not that deep. vllm comes with tradeoffs and IS better at scale, you do not have scale, why deal with the tradeoffs 
vllm and sglang have fancy batching and caching stuff that's good for serving LLMs at scale™
basically as soon as you have more than one concurrent request you probably want one of those
And the 2026 Nobel prize for least intelligible image goes too…
llama.cpp perfectly fine if you only have one concurrent request + much much simpler to set up

uu 
Batman Season One 1966 - Batman And The Tiger
uu
request that is with current
request that happens at the same time as another
multiple users using the LLM at once
this is a currant
ye what kni said
how do you not know what concurrent is

currant
currant?

it look like

bery
this is
black currants in dutch is liiteraly called "black berry"
sea buckthorn
i like white ones and red ones less so and the black ones are the worst
red black berry is aal berry
i immediately don't feel like using docker anymore

i dont remember where i've heard the chorus but i definitely remember hearing it
acquired taste <- ye, taste food (only if nose unlock)
acquired smell <- only if nose unlock
my friend doesnt like how it smells
i dont mind it
https://www.youtube.com/watch?v=I8aH0vPLdik
@real sierra i find cool ksp mod what do you think
Space Shuttle Challenger performs an extreme, uncertified contingency abort
For bug reports, open up an issue on the GitHub repo
System : i5 13600k@ 5.1 GHz , 64 Gb RAM, RTX3090Ti
KSP version 1.12.3
Modlist (non-exhaustive) :
-Real Solar system + dependencies
-EVE Raymarched Volumetrics (Early access)
-ballisticfox's RSS Reborn
-EVO (city li...
Provided to YouTube by Universal Music Group
HOT TO GO! · Chappell Roan
The Rise and Fall of a Midwest Princess
℗ 2023 KRA International Inc., under exclusive license to Island Records, a division of UMG Recordings, Inc.
Released on: 2023-09-22
Background Vocalist, Recording Engineer, Producer, Drumprogrammer, Composer Lyricist: Daniel N...
ye
do you actually need this or can you just allow full access to /dev and have everything work
i heard this in mcdonalds
had it open

not to /dev but specifically just > /dev/null
it works for everything
currant also good
i prefer black over red 
it does not
when i dont like an annoying app screaming about sudo i send it to the purgatory
can still see stderr 
if you dont like something then send it to hell
if dutch "zwarte bes" (black berry) is currant in englsih, then what is black berry from the phone?
just &>/dev/null then
ye, that's a bit better
but can probably still find a way to print to the terminal somehow
or do other bad things that would signal an error
actually, just don't execute it at all
nothing can go wrong then
just kill the init process

cant do anything then

pkill
I will start backreading from here 

i think you do yeah

kagamine len mentioned 22 times
Did you manually add all of the words to detect?

i lied you don't, but it does make life considerably easier so no reason not to
since when is len a profanity
Ahhhh
i like the name

we're improving our profanity with those one
its ranked now
Competitive swearing 
Fuck Docker all my homies use Nix
Also vllm works just fine on NixOS with CUDA
Already had it running without issues
Bro u don't know the stuff i have lmao
Release the ShAIdow files 
Stupid = profanity 
Nixpkgs package is broken currently
or at least on whatever version sam is using
Maybe not use unstable then ?
Idk last time i tried it worked fine
How much should i release

I mean you should release everything but I’ll be happy with a 1-3 sentence summary of all of ur cool forgotten projects
i made it multithreaded
Python multithreading
oh right 7900XTX


My favourite brand of GPU... 0x1002
cpu inference 
its only 187 lines, i need to up my profanity vocab
leper is profanity????
The term "leper" is considered outdated and offensive by many
oh
my bad
context is the game darkest dungeon
iirc
Guys I prompted him too hard
It’s been 18 minutes
He’s gonna run out of tokens
1 billion context length multimodal image/text/audio compressed 4b model, llm trained on ~150 million chat messages, vtuber llm, infinite Context video retrieval, some tts trained on my voice, some multilingual anime tts, llm trained to generate the human input instead of the response, audiovisual attention system to make a llm simulate focus, idk what else
I have a lot of old stuff
nixpkgs.overlays = [
(final: prev: {
pythonPackagesExtensions = prev.pythonPackagesExtensions ++ [
(python-final: python-prev: {
vllm = python-prev.vllm.overridePythonAttrs (oldAttrs: {
nativeBuildInputs = (oldAttrs.nativeBuildInputs or [ ]) ++ [
python-final.setuptools-scm
];
});
})
];
})
];
yes

not sure if this is enough of a fix
it needed setuptools-scm supposedly
its compiling rn so if it breaks it wasnt enough
ye, that's necessary when you're using an unsupported Python version
too new for vLLM
Why tf are u compiling vllm 😭 😭 😭
but that doesn't magically fix any other issues vLLM may have with that Python version

no docker good
oh
thes erver kicked into overdrive
the compiling is really going down now
Bro vllm is huge to compile
What is the specs of your server?
😭

it's fine, let him try 
failure is
necessary learning experience
2x xeon 6138. 256gb of ram, and a 4090
apparently the amd container toolkit simply doesn't exist inside nixpkgs
so that's fun

4090
the what

ye, the server was only 1850 euro. im probably selling the 4090 and profit
Bro just frigging make a venv and uv pip install vllm 😭 😭 😭 😭 😭
I would buy if I wasn't broke
Buy a rack or something at least 😭
don't really need a rack for a single server 
Cool
I have decided to stop backreading 
Its a bit crowded... but the racing sim and the 7900xtx laying there need to go downstairs
the empty boxes need to go in the attic
sam
pick a random number between 1 and 1382
947

why?
get back here, explain yourself
This is the amount for which you should sell me the 4090 for in usd
I know that's why
you rolled this
"A bit"
I don't even need a 4090 anyway lmao

that ayaya
"It's a bit crowded" 
the racing sim is 90% of the problem.
im not allowed to put shit in the attic rn cuz "they're cleaning it up"
i have waited for them to finish doing that for 2 years now tho
i recognize the charachter yes, but why does the number 947 corelate to that gif?
When my mom tells me i have a mess in gonna send her a photo of ur room
good question



Wait i just realized
You can literally run grok locally
I should make a @grok is this true bot
The other corners look better okay
ye the old ones
So basically this

Holy it's only 164gb
"only" 
Im gonna run grok locally
Bro i have 384gb
4x96gb?
3x 128gb
3 sticks 
Vram
oh
oh i see, okay
if tty1 is compiling something, can i go to tty2 and it will keep compiling?
ye

Chat it might be over today at work i accidentally blasted hyper energetic hatsune miku cover song
I told my team i listen to classical music lmao

why you lyin?
Bro is compiling vllm for every possible nvidia sm lmfao
For gtx 1080
Lmfao
For h100
it seems to be doing that yep
but im pretty sure blender has done that too in the past
@opaque sigil https://www.phoronix.com/news/NVIDIA-CUDA-Oxide-0.1

fuck you, nixos is great
the one called python3Packages.vllm

Concept render for new YouTube video player
gonna assume it's vLLM 0.11.2 then
if you plan on using the 4090 do yourself a favour and just use the damn docker container (after enabling gpu support and rebooting)
have fun 
already supports Python 3.13 apparently
so may work if lucky
Pov users who don't use docker:
Nice to see Nvidia still failing to build vllm like God intended
patch failed to apply huh
Docker does nothing but also break
And then be annoying because it's docker that's broken now
absolutely none of this
it either works on my machine or it doesnt work at all
that's charlie from the chocolate factory

cool i guess
that's the "i am a surgeon" guy
python environments 
Me UV
Still break
All break:-)
bro's running 24.11
ye
Oof it's over

i said ye cuz its true, it is 24.11
python3 -m venv .venv
source .venv/bin/activate
pip install uv wheel setuptools
uv pip install vllm
If only it were that easy
cuz thats what it installed

what
If ur running a trash os everything is hard

Wdym when I'm running 22.04 and 24.04 the only 2 supported os outside of rhel 
idk
Nix is a tool that takes a unique approach to package management and system configuration. Learn how to make reproducible, declarative and reliable systems.
here
i think
Live mother approach
did you go out of your way to go to the archive for older releases or what
i dont remember
i just grabbed the iso i used for my desktop and put it on the server


i'm honestly impressed
whats the issue?
aren't you using a flake there
do nix flake metadata
idk
(in flake directory)
ah
i feel like there has to be a misunderstanding here
does this dox or am i safe?
oh actually, im worng
the desktop is 25.11, not 24.11. i jsut misremembered the number
so thenw ehre does the iso come from?
unless you put those in your flake input urls for some reason 
but given that until 3 minutes ago you didn't know that you were using a flake I doubt it 
i found it on my cachyos drive
looks like i was actually correct, this is 24.11
i know im using a flake, i just didnt understand how its related
wait, so how does it being on 24.11 come into play?
for a flake you have to specify the Nixpkgs version in the input url in flake.nix
so whatever came with the iso doesn't really matter at all once switched to a flake
-# (except for the stateVersion but we don't talk about that)
Is good idea to avoid regardless of shadow
I’m officially launching a crowdfunding campaign to make this real
so yall say i can fix it by updating to nixos 25.11?
ancient unsupported version of Nixpkgs
apparently didn't have good compatibility with CUDA at the time it went EOL
and because EOL it didn't receive any fixes so it's still broken to this day
Who the hell is buying these
terraform + azure + managed aks kubernetes + helm templates + bastion + scim role provisioning🤤 🤤 🤤 🤤 🤤
- container registry
@wraith shuttle Rewrite message above in rust

@fathom glade rewrite this rewrite in c++
@molten valley rewrite this rewrite in python
unzip the llm and extract the stolen ip
@rich sedge summarise that rewrite and tell me how many letters are in "strawberry"
@spark flame
@vocal yew do a deep research and investigate whether this user has engaged in bad online behavior and write a report
@silent anvil ban him
<t:1778255769:R>
@old pendant turn all these inferior AIs into vedal soup
wtf you doing here

@whole mortar is this person credible
openrgb doesnt see my ram on my new mobo
Anyone wanna handle random production bugs for me pls
Wait this is literally genius
you mentioned your comapny handles banking info and is badly secured right?
why is there so much spam in this channel
Yes
what company is that? asking for no reason whatsoever 
Buy it idk
i got no monay
Move to the US and sue someone

Im gonna buy a 6090 when it comes out
Surely
It will have improvements
They cant lie anymore with less than fp4
So they need to actually improve the cards
im willing to bet a small amount of money on amd catching up
i think thats called the stock market
Fuck the msrp of 6090 is gonna be insane
a100 has dat fp1
excuse me, crypto?
Cause of memory
Cryptography
cryptography

Wait i couldve bought a 5090 with 3000$
ye
would still be in sm121 hell
cryptocurrancy
True
the guy i bought my new 7900xtx form bought a 5090 when it launched for 2200 euro
Buy rtx 6000 pro blackwell or go home
still sm120 sadge
But better
Wtf
vram is nice
download more ram
That's why i got 384gb
you know how people switch out the vram packages with ones double the capacity. could you do that with 4x or 8x?
packages wouild have to exist
for gddr7 yea
With 4gb in validation
that's why you have 9gb slopcards coming out
4090 has 12 modules of 2GB GDDR6X
4090 48gb mod
As of mid-2026, GDDR6X memory—which is produced exclusively by Micron—tops out at 2GB (16Gb) per module.

they use diff pcb with more spots
yea 3090
gddr6x and 5x are nvidia only slop so there's only whatever nvidia demanded
Literally what Chinese modders do
rumors of 3gb gddr7 apparently, with current max 2gb chips you're at 64gb
3090 can jsut go from 1gb to 2gb chips
3gb is out
4090 and 7900xtx oth need different boards
Gddr7 exists
3gb
Rtx 5050 9gb
thats not an april fools joke?

You know what would be cool
harvest billions of 5050s to upgrayyd 5090s
384gb hbm laptop
I can't believe toast cult is dying #sad
you need 4 5050s to upgrade 1 5090 to 96GB
no wait
math dont check out
you need 8?
why did my ability to do math go?
72
That one is good tbh
2 of em cost roughly like 1 6000 pro blackwell
you need 10.6666 5050s per 5090
Buy 1 rtx 6000 pro blackwell to upgrade 5090 to 96gb


you might aswell jsut use the rtx 6000 pro then
But it's not 5090
hmmm
Holy shit
harvest free broken 5050s from nvidia's factory's quality control
sam i think i have free time, if that unity build is reasonably sized i can test
<500mb
https://skymizer.ai/skymizer-announces-htx301-reinventing-on-prem-ai-inference/ vaporware that you surely didnt seem me link before '
Skymizer today announced HTX301, the first reference chip built on the HyperThought™ platform — redefining how enterprises deploy and scale AI inference. For the first time, ultra-large models can run on a single PCIe card. Powered by six HTX301 chips and 384GB of memory, enterprises can now execute 700B-parameter LLM inference locally at ju...
houston texas 301 
2010 sized indie game
its 100mb per release
2005 sized indie game
problem is discord max file size
is 500mb
Afaik it's only 600gb/s
do you not have nitro
According to reddit
i thought you. have
the cheap one
Sam if you were my little sister you'd have nitro


i can be



My sister gets free nitro from me
Expensive one
so many online accouints/subscriptions have a family member account that nobody uses
sam you know what to do
my sister isnt even on discord
the trials they give you to give out? i have them
Compact Design
- HyperThought™ employs advanced compression technology to minimize language model size.
Reduced Footprint
- Optimized for lower parameter counts and DRAM bandwidth requirements, utilizing standard LPDDR 4/5 memory.
Efficient Compression
- Weight (long-term memory) compression outperforms open-source llama.cpp by 9% to 17.8%.
- KV cache (short-term memory) compression with minimal perplexity loss (less than 0.06% to 3.52%).
Efficient.
Balanced Performance
- HyperThought™ achieves optimal area and compute efficiency.
High Throughput
- 100 GB/s bandwidth requires only 0.5 TOPs to achieve 30 tokens/second.
Robust Performance
- Even on a T28nm process, HyperThought™ delivers exceptional results.
Powerful.
Scalable Architecture
- Multi-core design for increased processing power.
High-Speed Processing
- Octa-Core HTX301 LPU achieves 240 tokens/second for Llama2 7B prefill.
Multi-Chip Scalability
- Connect multiple chips to achieve extreme performance – up to 1200 tokens/second for Llama 7B prefill and support for 600B large models.
Secure.
LISA v3 Architecture
- Incorporates Language Instruction Set Architecture (LISA v3) for enhanced security.
Secure Design
- Protects every interaction with security-focused instruction sets.```
their claims are le epic
the what?
but its fine, we already submitted the files and i doubt the profs will use a mac
ok
i have never seen that before in my life
i know
i dont think they get gift
it works 
this is just because of the tautology
can't say i'm the biggest fan of them vomiting .ll and .ptx files into the root of the project but oh well
at least they didnt yeet them into /home/$USER
vibecoded 
their vibe: fk u
quite the painless devShell thankfully 
Sam if you're my sister You're gonna have to endure this
Bro she cosplays animes and shit😭

japonslop
Her discord profile is full of oshi no ko shit
Last time she extorted me into watching the amazing digital circus while i was in the psych ward
based
And that's how i got into it
Nah
You don't have extortion material

Like my sister does
Ok i should probably reframe the previous statement
Wait what
djsfhskdjhf
i hate how btop shows 55% usage when its just using all cores but its pinned per core and not using 32 threads
@olive sable you don't have the picture where i pose like a femboy in the google corporate office elevator
i have the one where you did that in an elevator in the google office
whatd oes posing like a femboy even mean
yes
Holy fuck
do i not get to know 
i also have the files for the lilac 3d model i guess, but i dont really want to actually extort you
Lilac 3d model idc
model me like one of ur fr*nch femboys
I think I’m gonna try a 750k parameter LLM
Cause if it works
I can likely get something coherent out of 175k
And thus have something 1,000,000 times smaller than original ChatGPT
base it on quadtrix https://github.com/Eamon2009/Quadtrix.cpp
AI engine built in C++ /python to run Language Models directly on your own computer. - Eamon2009/Quadtrix.cpp
can we save this for when shiro’s in transit 
it is in fact a cool model

I just want to see the reactions of total strangers when shiro is in a gc and someone sends a tposing girl
difference in base pos
A pos is better
cuz stretching
you can see how the T pose girl has no shoulders
Sam irl face reveal

Would you rather I send THE picture again? 
may i introduce you to
1:4 sparsity
-# how disappointed you'd be if i leak it abit i wonder
looks like a rigging issue
you have a bigger issue with elbows and knees
but you just do the right rigging techniques and you chill
its an exaggerated image
as long as you got enough polys it wont look too bad
Is 10000 enough?
there is also the technique of modelling t-pose as it's easier
then changing to A pose and correcting so it looks right
i personally feel like jsut modelling in A pose is easier overall
Y pose. M pose. C pose. Then A pose
you don't get to align things to axis
I modelled in both
and do find T pose a lot easier
I think modeling is more of an art form than exact number
So things not aligning to axis is not always an issue
if i model in ais aligned things get too rigid
big dog
"oh no my arm is too long!"
way easier to you know
scale it on the axis
you can ofc still scale on axis when it's diagonal
but it's more clicks
what yall doing now?
So in short, skill issue 
who on earth was reporting harassment?
proportional editing 😍
Me, I'm reporting @olive sable 
Why?
it's nice but not always what you need
wth?
how many toasts are left?
WHat ARe WE DOing?
actually, there's still a lot

Wtf
8 total, including the real toast
Discord lie
what song u listening to?
crack
Turning sporadic groups into puzzle
I've always wanted to understand Sporadic Groups better, so I designed some physical twisty puzzles (like the Rubik's cube) to explore them and get a feel for how they work. I couldn't be happier with the results.
*Minor Correction: My definition for normal subgroups in the video was a bit too strict. It's only important that this identity hold...
who
you
then why mouse freeze?
Dawg I was talkin' bout a meme

Now I changed song 
https://youtu.be/4soZ33MvlW4
Why rendering a single letter on your screen is one of the hardest problems your computer solves every day - a deep dive into Bézier curves, rasterization, hinting bytecode, subpixel rendering, and the GPU texture atlas cache.
0:00 The mathematical nightmare behind a single keystroke
0:33 Fonts aren't pictures — they're Bézier curves ...
ik its random but does anyone have curly hair? 3b
nope
I can only offer curly bracket
https://youtube.com/shorts/Mo9uh3wdH5I
This is a snippet from my video on the C Iceberg. Here I demonstrate C's "digraphs" and "trigraphs," which let you replace curly and square brackets with combinations of other characters.
why tho?
Someone on the comment actually submitted their C assignment using digraph
i thought id ask sm1 that why on earth are my hair so oily even after i wash it thoroughly
Should we call the Ghostbusters? 
why tho
i'm looking to ban you 
who's looking to ban who?
hello moderator person
cloud, serve rmod, banned the guy asking ofr a SOL wallet
phantem wallet?
some crypto BS
oh
wtf even is a phantom wallet
Who knows lol
website it seems
is it a wallet that lives on those flying mc creatures that attack you if you don't sleep
dam
I forgot they are called phantom
not too far off
MODS
does anyone know esli mendez?
Sam in reversed is mas, which is a pronoun for older male 
yes
its :small_red_triangle:
that's fine then 
😭 wdym older?
Well you call someone that you thinks are older than you "mas" in Javanese
as long as it isn't over 10mb you're good
does anyone know esli mendez?
That's insane
im only 18 tho
What file extensions are they even accepting
this isn't a topic for this channel, please stay on topic or move to general-network
i think any
dam
but whats the topic?
This is one of those rare moment where I wanted to try it 
read the channel name and topic/description
this is #programming
dam

Yeah, and if you are being talked to by someone younger they will call you mas here. But nowadays it is used to refer to generic young man, not necessarily older 
Language evolved too after all
Even C language
i feel like this takes new user hate a bit too far 
Like the video that I sent. Who knows C has a digraph
its compiling rust now 
firefox compiling rust 
what are you doing 
"Everything is precompiled anyway" oh how the turns table
Try wd 40 or CLR
building firefox it looks like





have the server just laying on a random shelf










