#programming
1 messages Β· Page 109 of 1
lmfao
9GB from what I can tell
i though it was 90gb at first but 9gb or 90gb both are problematic
What kind of audio model and on how much data, using what software?
im using process explorer
And what params?
yippe
Can it tell you GPU VRAM usage, CUDA usage and total CPU usage?
If not, it's useless here
while running the ai?
Yes
if you're on a relatively recent windows build you can do an in-place reinstall, sounds like you need it
Especially we should be looking at the GPU VRAM usage
Yeah, or just nuke it and go Linux like me when my Windows broke
when i said hi my gpu went to 13% and vram was at 4
speech tokenizer, emilia dataset(streaming) on batch size 32, Flax NNX. Following DAC recipe
This should have less than 100M params total including discriminators
Oh also to answer this yes. I only have 16GB and this on AMD. And apparently my card isn't compatible with flash yet so i can't even use that.
is your taskmgr overridden in the image file execution option
there are like much more issues
So you're most likely not using the VRAM
awa
wait a second
i dont know if i have pytorch cuda installed
Speech tokenizer, huh
Interesting
Try different optimizers maybe, that does make a different in VRAM use
what's the output of gi 'Registry::HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution Options\taskmgr.exe' in powershell
Recommendation: figure out how to allocate stuff to your GPU using your LLM library of choice, then test how that effects VRAM usage
PS C:\WINDOWS\system32> gi 'Registry::HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution Options\taskmgr.exe'
gi : Cannot find path 'HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution
Options\taskmgr.exe' because it does not exist.
At line:1 char:1
- gi 'Registry::HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\Curren ...
-
+ CategoryInfo : ObjectNotFound: (HKEY_LOCAL_MACH...ons\taskmgr.exe:String) [Get-Item], ItemNotFoundExcep tion + FullyQualifiedErrorId : PathNotFound,Microsoft.PowerShell.Commands.GetItemCommand
Note that you need to close any graphically intensive apps to get an accurate result
i only have youtube and discord open
YouTube does use some VRAM, so you may want to close that once you get the models on GPU
its audio only...
does taskmgr start at all when you observe the process list in procexp
Otherwise you will run out of VRAM and your performance will die
The browser itself uses VRAM
And YouTube code is garbage pasta code
nothing the name taskmgr isn't found in procexp
msiexec.exe
but thats for running .msi files
and then some python from eariler on today
you're gonna need to start the process with a debugger to tell why it's dying then
ok
no error just insta closed
1
C:\Windows\System32>SFC /SCANNOW
Beginning system scan. This process will take some time.
Beginning verification phase of system scan.
Verification 100% complete.
Windows Resource Protection did not find any integrity violations.
C:\Windows\System32>

Welcome to The Swarm FM, 24/7 neuro radio where you can listen to music from Neuro-Sama and Evil's karaoke streams to your heart's content. Remember, watchi...
something I've been making for the past 5 years
the project: https://github.com/9001/copyparty/
watch this video on the demoserver: https://a.ocv.me/pub/demo/ (try showcase-lq.mp4 if your device cannot play the hq.webm)
Gotta love it when I am paying for something through PayPal and then right after my transaction succeeds, the website glitches and now I actually didn't get what I purchased, because PayPal couldn't reach back to the merchant
(venv) C:\AI\NeuroSama\rvc-cli>python rvc_cli.py infer --input_path audio.mp3 --pth_path neuro.pth --index_path neuro.index --output_path output.wav
Using HiFi-GAN vocoder
Converting audio 'audio.mp3'...
Conversion completed at 'output.wav' in 6.29 seconds.
they give you the hard capchas now i once spent 15 minutes on one
All because I switched browsers too π
HiFiGAN was quite good?
@small anvil how's nxlang doing?
im reworking the bytecode spec to be more coherent
and then after that ill create the vm that actually runs it, then ill create the compiler
and then hopefully after that I can integrate JIT
or just an actual native compiler
that's certainly a work load
I am awake 
congrats 
Thanks 
Instead of going to bed late and waking up late, im now going to bed early and waking up early
Pitch too low, try upping by 12 semitones
ok
Agreed.
i did but it sounds like the same
If it sounds the same you didn't increase the pitch
[Original AI Reply] Wait, what?! New test?! I swear, every time I think I'm free, you come back with another test! [Final Sent Reply] Wait, what?! New test?! I swear, every time I think I'm free, you come back with another test! [RVC] Running command: C:\AI\NeuroSama\rvc-cli\venv\Scripts\python.exe C:\AI\NeuroSama\rvc-cli\rvc_cli.py infer --input_path C:\AI\NeuroSama\rvc-cli\tts_audio\1399575658941649037_reply.wav --pth_path C:\AI\NeuroSama\rvc-cli\neuro.pth --index_path C:\AI\NeuroSama\rvc-cli\neuro.index --output_path C:\AI\NeuroSama\rvc-cli\tts_audio\1399575658941649037_converted.wav --pitch_shift_semitones 12
I'm just guessing, but maybe either there's something wrong with the inference engine in use, or your argument should be in a different spot
Or you used a wrong argument, that is also an option
Are you able to download NeuroSynth yet or not?
NeuroSynth is my thing, but isn't even applicable to your usecase
is anyone going to be able to download it>
im doing gpu stuff now 
as expected i have a single physical device that supports vulkan, since i dont have an igpu
My internet just went out :(
welp, AI reproduction lol. Or should AI spreading "weights" should be reproduction?
https://www.youtube.com/watch?v=4b4S-duf0sw
ASI-Arch autonomously designs new top AI models. #ai #ainews #agi #singularity
Thanks to Hailuo for sponsoring this video. Try Hailuo 02 today! https://bit.ly/hailuo2
AlphaGo Moment for Model Architecture Discovery: https://github.com/GAIR-NLP/ASI-Arch
0:00 Background of AI innovation
2:26 Previous AI methods
3:35 ASI-Arch autonomous researc...
did someone say massive?
you know what else is massive?
the eiffel tower
but what else?
my debt
okay shit got too real

kurzgesagt said that'd a evil AI would spread itself via sharing its weights, so, would a LLM or whatever would replace LLMs, is basically its weights?
https://www.youtube.com/watch?v=uiPhOk1t3GU
wait, it wasn't them, but is very similar.
In this video, we walk you through a plausible scenario in which AI could lead to humanityβs extinction. There are many alternative possibilities, but this time we focus on superhuman AIs developing misaligned personas, similar to how Microsoftβs Bing Chat developed the misaligned βSydneyβ persona shortly after its release.
ββββ...
what
don't think it is anything more than a probability based math calculation
it doesn't even have randomness inside it
let me see if i can find the timestamp.
In this video, we walk you through a plausible scenario in which AI could lead to humanityβs extinction. There are many alternative possibilities, but this time we focus on superhuman AIs developing misaligned personas, similar to how Microsoftβs Bing Chat developed the misaligned βSydneyβ persona shortly after its release.
ββββ...
if you have a debt it might not go well soonβ’
we are one step away from a bingo being filled
researchers already gave the model tools so it could do that
and persona to keep itself
so it is nothing special tbh
Because i think someone at some point told me that a LLM or whatever one of its variations, spreading itself like a virus, as fiction portrays, would be complex. As in, making perfect replicas of itself through the internet.
okay im sorry sam from pulling up brainrot memes
π
oops, ran linter too early
my git commit is fucked but who cares
man
I have no internet so I have to use my phone internet π
awa virus
awa
ok i have made the vulkan physical device and logical device code 
now all i need to do is make the window surface, the swap chain, the image views, the shader modules, the render passes, the framebuffers, the command buffers, the frames in flight, the vertex buffers, the index buffers, the uniform handling, thew texture mapping, the depth buffer, and the multisampling

it looks like it would be best suited for a hoodie as it requires 2 layers
gnome terminal just crashed again 
trying to debug my segfault generator, but gnome gets in the way 
segfault generator
generates segfaults
not on purpose tho
handwriting machine code really is something else
day 2 of writing a program that prints file size
wow thanks YouTube for introducing the GeForce GTX 680 13 years later!
awa
awa
rain mentioned??
I, as a humble inabakumori enjoyer, cannot let the word 'ame' slide
it has a lot actually
though i guess its mostly the same songs mapped all over again
let me open my osu lazer in the middle of night
there are 14 different songs according to my possibly incorrect calculations
and not counting non-std gamemodes
19 including remixes and colabs(except that one massive collab where inaba appears for like 10 seconds)
16 w/out colab
fair
but i suck at this game tbh
where do you even find that emoji
neurooper

staz
neuro >> vedal ahh statement
shit, gdb really loves crashing my terminal
well did you touch some grass or not?
I just did rolling window event cluster analysis to find times of interest.. and now I gotta dig through 1.2GB of log files. someone help me holy fuck
am a programmer, not a data scientist 
i've had to deal with similar issues when developing networking software...
Streamed Live on Twitch: https://twitch.tv/tsoding
Enable Subtitles for Twitch Chat
Chapters:
- 00:00:00 - Intro
- 00:02:12 - Low Effort TTS-ing of the Article
- 00:10:12 - node.h
- 00:12:15 - Starting writing cood
- 00:18:17 - Empty addon
- 00:25:06 - Making an .so
- 00:26:43 - require()-ing the .so
- 00:32:41 - node-gyp
- 00:37:01 - Exporting...
Why? QwQ
Node haunts ne in my dreams... please don't make its use cases even bigger
You have unlocked new role
Yeay I guess? XD
All I'm seeing is just Node-API Node-API Node-API Node-API
There's a reason it's usually just called napi
how do I pronounce NAPI
nappy 


fr fr
bakushin
This is the shit I'm dealing with lol.
I don't have any logs as they were rotated away, but I do see these events here (dropped files that are not supposed to be dropped). Turns out that version before 3.6.3 had a bug with this exact thing...
But at least the mystery is solved 
Only took me writing a god damn 400 lines shell script, various database extractions and a lot of dying brain cells :D
digging in log histories to see what your software did at a customers installation? cause it sucks ass
and the DB we got was corrupted lol
the cables under my desk
let's see how well is your server

heya #programming
i was wondering... is plain js better than js libraries in terms of websites/frontend ?
if so, why? π€ i just wanna ask cus im considering going thru them later
hmmmm something that doesnt take a toll on time like npm run dev just to test a page
(experienced with next.js app for vercel)
i am here
i mean.. once npm run dev starts you can hotload the website no?

like how live server works on vscode?
stream your vulkan dev journey in voice chat 
no
aww
basically yea. i think most stacks support some kind of hotloading nowadays
but it also really depends on what you are building
90% of vulkan deving is copy pasting code form the tutorial and reading wha it does and how it works
if you want to build a basic blog with static elements.. don't even bother doing any JS. just use jenkins or something (idk what the current tools are)
π€ i see...
i find plain js to lack features i sometimes need
<--- astro glazer || or use astro
no, plain JS is ass because you have to directly mutate the DOM
i just wanna know if they are any faster tho
right.. astro
I mean for a basic blog, it works
js runs js. unless you compile your website to wasm and just do that. it will run the same
just make your website in unity 
write it in Windows Forms and spin up a VM for every user
also, what's a .ex
Elixir source code file extension
ah

???
like my screens literaly look like this
I'm proud I could give you a mini heart attack 
if I had multiple monitors for myself, would have it set like this as well
for now 
but yeah there's a lot of boilerplate that will likely never change because you're just not using those Vulkan features
woah triple monitors?!!!
rich

i got 1 "brand new" cuz it got returned tot he store for 270.
and 2 came 2ndhand for 300 combined
thats normal
is that usd
???

yes it is, you're not gonna put the code for picking the physical device in a different file from picking the logical device
this is only 338 lines so far
I'll do PhysicalDevice and LogicalDevice, extending from the Device class 
hong kong?
yep
huh
you're gonna do what?
surely those aren't completely different, incompatible concepts that just happen to both have "device" in the name


338 still sounds fine
I'd maybe separate it if it grows too much though 
i think id eventuially end up with maybe 4 files. but im not gonna split it yet cuz i dont know the api well enough
ill follow the tutorial first
oh
now I see

I know nothing about vulkan don't scold me

you do need the physical device defined before making the logical device
three similar monitors strat
its 3x the exact same monitor
besides 1 of them being newer so slightly diffrent colours
but why? it sounds like a waste of space
how is that different from js libraries like react or next.js
i thought it was the opposite
that's the joke
I'm still sticking to my single screen
so 4 is kinda the max I can have, physically as well
rich
my max if do cross gpu stuff is probably 9, but id have to use tv's to fill that cap
tbh i havent had much issues before with running 1 display of the igpu on my old pc
so it should just work fine i think
azaka works with windows professionally
Oooo
dwm eats up a huge chunk when you're trying to render something on gpu 1 and displaying it on GPU 2
I've been working with a 12" thinkpad for so long I'm just used to not having huge screens 

i do enjoy having multiple monitors
my laptop is already screen heavy, it would just tople over
I'll be honest 200 dollars doesn't sound ridiculous
and 30 days warrenty
im not expecting the screens to be great at that price tho
chat im sorry if this is a hard ask i just need help ig
so excel has code right?
VB something
when you're traveling and having to work I think screen's quality would be least of the concern
i wanna be able to copy paste this inside a text box in excel
i guess
excel formats it for me
and then puts it all inside cells
in order
my question is can i use javascript for that
sorry i dont use excel
afaik macros are forcefully disabled now except signed macros
nvm only those with mark of the web
This might take a little bit
opus
it's a 30 minute recording of a lil dj set i did lol
It's technically still converting
γγ€γͺγΌ γγ€γͺγΌ γγ€γͺγΌ
γγ€γͺγΌ γγ€γͺγΌ γγ€γͺγΌ
γγ€γͺγΌ γγ€γͺγΌ γγ€γγ€γͺγΌ
γγ€γͺγΌ γγ€γͺγΌ γγ€γͺγΌ
γγ€γͺγΌ γγ€γͺγΌ γγ€γͺγΌ
γγ€γͺγΌ γγ€γͺγΌ γγ€γγ€γͺγΌ
β1ζιθδΉ
γ―γγ‘γ
https://youtu.be/QSFVBqj2-Lg
#γ¦γε¨
#γγ¦γ«γ€γγ€γͺγΌ
If I got Linux to be able to use both my iGPU and dGPUs at the same time I could use up to 9 monitors, my motherboard only appears to have 1 display connector, both dGPUs have 4 each
But not like I could afford any more monitors or even fit any more than the 2 I have
i would have to put my old cpuin since my 5950x doesnt have an igpu, then id also need to put my old gpu in for those extra 4 ports
for displays i have 7 pc monitors in my house, and the rest would have to be tv's
we have a lot of tv's here so should be fine
i just cant be bothered to do all that
Well, I don't even have enough monitors
a large part of the job of something like React is to keep the DOM synchronized with your application state
It also does so in a way that makes XSS vulnerabilities really hard to introduce
Plain JS fails hard at both of this
oh
vulkan?
wait i can't switch it over to the AMD card lmao
it must have either picked the first one that fits the game's needs, or it gave the 3080 a higher score for some reason
The game must really like Nvidia
i've never had a game prioritize my 3080 over the 7090xtx before

Must be a badly made game
it all depends on how they do their vulkan choose physical device code
i work with a score
You should try to make it default to the device running the main display
And if that is an integrated GPU, check for dedicated GPUs and run on that if found
so what you're saying is just run on dedicated if there is one
which is what im doing already
Do be aware of systems with multiple dedicated GPUs like mine
What would your code do in that case?
fixed 
it would pick the one with the highest score
And how would it compute that? Because I don't want games running on my 3090 which would mean the game has to stream its images trough a chipset PCIe x4 slot

you can basicly choose how it calculates that, but rn its just 0 points for whatever doesnt have the needed capabilities, and for those that do its just the max image dimensions + (1000 if its a dedicated gpu)
you could always just let the user decide
You should add a significant amount to the score if the device is connected to a display
thats not really exposed directly in the api i think
Valve mostly fixed this in Steam by having fancy shader cache tech 
Well, it would be good to at least try to make it so a device that isn't meant to be used for games isn't used for games
Can you somehow compile a small Linux executable I could use to check which GPU it would decide on on my system?
And print out the numbers it uses to decide that?
dunno why it doesn't work in this case, Vulkan games should be well supported by that
skipped the shader cache preparation maybe 
use vulkaninfo
and just read the properties from that
You're writing C++, right? Just make a small program with the score calculations that outputs to the terminal, and compile that
it would compile to just an exe tho
Then change the compile target
-#
this tool already exists
The output is like a billion things, how am I meant to find what I'm looking for?
but like, how?
It depends on the IDE and build tool
maybe don't listen to him on this, C++ cross compilation is a massive pain
it would be easier to send the source code 
why doesn't this work? C:\AI\NeuroSama\rvc-cli\venv\Scripts\python.exe C:\AI\NeuroSama\rvc-cli\rvc_cli.py infer --input_path C:\AI\NeuroSama\rvc-cli\tts_audio\1399575658941649037_reply.wav --pth_path C:\AI\NeuroSama\rvc-cli\neuro.pth --index_path C:\AI\NeuroSama\rvc-cli\neuro.index --output_path C:\AI\NeuroSama\rvc-cli\tts_audio\1399575658941649037_converted.wav --pitch_shift_semitones 12 it still sounds like evil
Whuh
Did the pitch change at all?
sth sth zig fixes this 
Doesn't sound any higher
Then your pitch shift is not correctly being applied to the RVC inference or your inference engine sucks
it actually does 
a lot of languages do fix that
except C++ because it's old and bad
How strange
No I meant zig fixes this for c++ since you can use it as a compiler with musl
How will Sam build multiplatform builds then, last I checked Vulkan doesn't run in a browser
it has a full C++ parser? or does it rely on LLVM?
Yeah it just passes things to clang for the most part
have fun ig
There's some differences with the argument parsing iirc and well, musl obviously
But most things should just be passed as is
whateven is the support for webgpu nowadays. Last time I checked it sat at 70%
I don't think I can compile this as-is
Just give me a Windows exe I guess and I'll see if WINE works with Vulkan stuff
I think chromium supports it but it's nightly only on Firefox or sth
Well the main parts
Firefox is working on stabilizing it on both Windows and Linux over the coming weeks and months
le epic totaly safe exe
SMOL
they're posting weird executables in chat again
I don't think it'll run on WINE
time to play dynamic linkage bingo
output shoudl looks omewhat liek this
oh yeah, that's not even the Windows thing
Discord mobile didn't load the image so I assumed 
3 kilobytes smaller

huge gains
thats what it does yes
It did correctly find the 4070Ti, so fine for now
the thing its supposed to do for seeign which gpu got sleected is just a print

i havent finished doing the triangle yet
Since you're doing native Vulkan now you may want to figure out how to get Linux builds at some point, either via figuring out how I can compile it for Linux targets or figuring out cross-compilation yourself
btw, if you want to do GPU selection in a smarter way later you can check https://vulkan.gpuinfo.org/
they list all the properties and stuff for the various devices
so you can check which it would pick without needing the hardware
uhhh
i was thinking of just saying "use the web build" for now
There isn't exactly Vulkan on web
or someone on linux compiles one on their pc
also this thing probably
https://www.lunarg.com/wp-content/uploads/2021/09/Enhanced-Devsim-15Sept2021.pdf
i tried compiling sam's file for windows on arch out of curiosity and zig was NOT happy 
Well, as you may or may not know, I have a device running Linux called "my PC"
ye but do you have the vulkan sdk?
Linux SDK?
and c++ stuff too ig
i meant vulkan
I doubt it'd be hard to just apt install it
my brain likes crossing wires
So what dependencies and build system are needed?
i just use makefiles, and you basicly jsut need the vulkan sdk since there you get the option to install sdl2 and glm with it
gnu make, gcc or clang, vulkan, thats probably it
you can just compile it and see if it complains
and download the missing packages
i use clang but it will probably work with most c++ compilers
why not just let a CI do that for you instead of doing manual builds for every target
because now you have to involve ci 
sure but what are the odds of your version control not to have some kind of access to it
what did it complain about? everythin im doing rn should be cross platform compatible.
all i cant hink of is you not having the vulkan sdk so you dont have the validation layer file.
windows stuff
you can just disable that
cross compiling to windows isn't as convenient sadly since you can't rely on musl
i dont even know what that is
but there should be no platform specific stuff hapening
Me being on Linux and trying to produce a windows executable
glfw handles that under the hood suposedly
oh
a windows exe?
but why?
did the linux one not work?
Oh it did
huh
Vulkan SDK has been obtained, GCC was already installed with something else
i wanted to output a Linux binary on windows but accidentally booted into arch so I tried the other way
Now where can I find something I can try to build?
You need glfw3 too btw
this list is cool but idk if its realistic to check all of these
im pretty sure vulkan just has a performance check function
GLFW installed
APT is so good
It just has all this stuff conveniently available
Yeah just 3 year old versions of it 
you could probably automate it, but you wouldn't even need to
the point is just that it allows you to see the properties (and maybe more importantly, extensions) without actually owning the device

So what now with all that stuff obtained?
honestly just prefer the first GPU and let people that have multiple set their preferred one via env variables 
this website is kidna the same as this program that came with the vulkan sdk
the website is more liek a database tho
The package manager

Btw if you're using vanilla apt there's also nala as a nicer frontend for apt with parallel downloads
apt is cool until ubuntu decides to give a shit on what you want and use snap in the background instead
Well, I'm on Mint
Do they still do this?
they do
Cool
Package chromium-browser
plucky (25.04) (web): Transitional package - chromium-browser -> chromium snap [universe]
2:1snap1-0ubuntu3: amd64 arm64 armhf
snaps 
canonical and redhat both belong in the bin
I get canonical but why redhat
hm

is it a good idea to change a manual process I run after every karaoke (never late surely) into a github action so I don't forget to do it
(this will involve mostly rewriting said process so it can work on an action without imploding)
because the stunt they pulled with centos. They cut changed the end of life from 2029 to 2021 out of nowhere and then just discontinued it altogether in favor of stream
after it was already released mind you
What does it even do?
like shazam but for neuro stream intros
I need to keep a db of what songs play, and build a file from that
atm I'm manually generating that file
How silly
oh yeah, that whole thing
I kinda forgot since Rocky Linux and Alma Linux have replaced those builds
get rid of them both
We moved to alma at work as well
yeah sure I'll write an action for it, what's the worst that could happen
too bad that there arent too many stable and well run linux distros out there I would use for production
@olive sable Well, if you want to figure out if I can compile Linux builds, let me know or something like that
sure
did you actually try to compile it yet
I don't have all the files yet
I think I lack a build script and didn't download the code yet
isn't it just the main.cpp? it has, like, no dependencies
won't even need a Makefile at that point
Well, I guess I'll check if I can just build it as-is
my makefile would probably be harder to run than just clang++ main.cpp -o a or whatever you do on linux
Is this meant to happen?
uh lemme check
Good old implicit declarations
just do include cstring i guess?
i wasnt including it apprently and it worked fine
it even says that it came form there
You must be including it somewhere
yeah, probably because some other header you include can then include the cstring header
so it will also end up included in your source code
What even happened here
looks like it couldnt find the vulkan stuff i guess?
that or some C++ version or compiler specific issue where certain headers are always included? I think I remember that being a thing
-lglfw
They do that with libc but I don't think it should happen with c++
c++ mode is a lot stricter
I added that but nothing happened
Gimme a bit I just got home
Or look at Sam's repo for the cpp game engine for some flags
There's some stuff missing that should be linked with
this isnt in that repo, its a seperate vulkan project file
Well, what compile flags are you using on that one?
the amkefile looks liek this if that helps?
some of those arent necesary
wait
i basicly just copied my makefile from the other project for this
some of those ldflags are platform specific i think and probably not necessary
hes using gcc iirc
Yeah don't do gcc do g++
Yeah you at the very least need that
There's also uhhh
Some missing references to some triangle hello world?
Oh yeah I can't read
looks like -lvulkan -lglfw is enough on my end right now

doesn't display shit but it runs
Don't use gcc use g++
3/4ths of what vulkan needs to dispaly stuff is missing
Oh, GCC doesn't work for this?
shouldn't it at least open a window
gcc is the c compiler, you want the c++ compiler
Oh cool, it works now
ye it should
glfw supposed to make one
I'm sure it's encrypting my files as we speak


Interesting silliness
the loop is probably saturating your cpu cuz it has no regulation at all
like its probably runnign at multiple thousands of times per second
no window so no vsync
imagine using vsync
oh, it's for the events 
About what I found out from looking at the code, it fully saturates an entire thread
it sure does
normally you'd have code to stop that, but this wasnt exactly meant to be a finished product
i have:
the physical and logical device selector, the window surface
i still need to make:
the swap chain, the image views, the shader modules, the render passes, the framebuffers, the command buffers, the frames in flight, the vertex buffers, the index buffers, the uniform handling, the texture mapping, the depth buffer, and the multisampling
oh, it doesn't even set up the swap chain yet
makes sense why I don't get a window then
It does for me as well
ah
probably a compositor thing 
Probably something specific to what Shuni is using, yeah
I assume Hyprland doesn't actually display a surface until it presented a frame
-# probably wlroots in general?
Now that that's figured out, you can have native Linux builds I guess
idk why but i kind of expected it to blow up 

I have awoken

oh yeah now that i'm home i can try cross compiling for linux
wait no
it still depends on vulkan and glfw
And Sam said he didn't want to run benchmark on first run 
there is nothing to benchmark yet
ofcourse it will run at 10000fps rn, we're barely touching the gpu
fps, more like events per second
since thats all it does rn
a wa
perhaps




i think ninja is not too happy 

Don't worry, I am not happy either 
segfault btw
Because now I actually need to git clone Sam repo to properly edit all the bracket
Good idea

i will still repond to pr's with a no effort "no"
Wdym no effort? I have to learn regex all over again everytime
Yes, because that is what contributions are. I am providing you with something so you don't need to do it yourself 
I think they can just push new commits, but that might also include force pushing
worst case you can give it an SSH key that can force push 
and i will put no effort into reading what you send and jsut do "no"
I've pinned down the trigger conditions for a small bug I've been having on discord the last few days.
Is it something cursed?
No just Phantom new post dots when a channel meets certain criteria.
If the last post is an addendum to a reply close enough after that it's merged and I close and reopen discord the channel will say there's a new post.
Hi chat, time to disconnect from this wifi and have no cell service
Are you going camping?
Just realized the last post in one of the affected channels isn't a reply just an addendum.
Acerola just dropped another vid
The stop killing games situation is crazy -- sign the petition https://www.stopkillinggames.com/
Bungie Voice Actor -- @mokadevs
Support me on Patreon!
https://www.patreon.com/acerola_t
Socials:
Twitter: https://twitter.com/Acerola_t
Twitch: https://www.twitch.tv/acerola_t
Discord: https://discord.gg/FxGQvbfm6Y
Thanks for watching!
This vi...
So I guess it just needs to be a merged addendum
1 hr long lmao
not the absolute collapse of the Ga
Have you seen his comment? The dude is needing help in Uma Musume lmao
elf sword
morn
hi
Elf word should be named wisdom
i read the entirely of hobbit, lotr and silmarillion on a 128x128 phone screen
why
Because chay love cursed stuff is my best theory
i have these books in hardback, its important for life
Well well well
would you look at the time
its bot-o-clock
@rare bridge
sad
get a bot with embed perms idiot
execution bime
hi scammer
also in #gaming and #artist-alley
it seems to be always the same ad, so it has to be the same person running the bot?
still promoting the instant ban of anyone who sends 4 images
More like same company

is discord scamming that profitable that there are entire companies for it now?
different stolen account each time
nah its probably just a RAT
Scamming people in general is profitable. 0 initial cost (or very little) with high gains
you'd think that most people are smart enought to realize its not an actual mrbeast challenge or whatever
Most people are indeed
the key is doing it at scale
The problem is that you don't have to get most people
ah
And also, the more outrageous the thing is, you are filtering those that are the most gullible only ironically
yeah, that's why it's spammed in every channel on large Discord servers
sooner or later someone will forget to engage their critical thinking and will be taken advantage of 
I feel so dirty creating a Option<Box<[Box<[T]>]>>
sad
you should
cant they just freeload off the taxpayers like everyone else?
its ok Box is a fat ptr so Option<Box<T>> is probably cost free
i hope 

I think? it should be better than just having an empty boxed slice since that still triggers a heap allocation even for zero size
yeah
im curious why the extra box though
I have big brain moment. Linked list is just a tree with 1 branch
cuz they are nested lists. Sure I can use a flat slice but then I would have to track the start and end points for each
last element points to first element, suddenly your program never terminates

@tender river cursed binary modification idea
function ptr inlining
i did think about something like that
My hypothesis is once again proven correct
having a function that fetches a function pointer on initial run and then replaces its own code with a jump to that pointer
anything more than that would be a bit hard
i was thinking creating a function that copies the whole binary into ram and grafts the function into the binary
thats gotta be the best pirate biggest struct ive ever seen 
I'm hungry
im konii
Hi Konii
im dying
it can be a regular linked list or it can be a list with cycles, like it can be a tree or a directed graph
isn't that basically just dispatching like you would do for different CPU featuresets
i'm hungry too but my potatoes just finished baking so i'll eat them in a bit
for portable simd, sometimes. other times its just a branch on detected featureset bits.
Don't have oven here. Sadge
typically would depend on hot vs cold path
you can use a fire
but you'd need foil
or something similar
grilled patatoes sounds nice
Lemme just ignite some in the middle of living room
get a microwave and take out its transformer

then use the transformer to power a lazer
Baked potatoes with butter and herbs indeed sounds nice tho
then cook your food with the lazer
as a person who knows a bit of electronic i dont reccomend this.
as a person who is hungry i fully endorse this if its to make food
yep
just popped in and thought you were talking about ml
I know what I am doing and I don't even wanna deal with them lmao
dead before you hit the floor 
ML brainrot
ive only touched big transformers twice or so i think. only 1 was on purpose but the devices were all discharged
not discharged as in a couple hours, more like 3 years
advice is now shown to be solid for 3 distinct cases
Transformer doesn't really store energy btw. It's the caps that you should be wary about even when it's off
as a corollary it can be shown that i am not dangerous as i do not contain a transformer

You have something worse. Electron gun
i had an rgb lamp with tasmota that i really liked but the capacitor died 
i have the replacement but i'm too lazy to solder it
please call it properly
cathode ray tube

also i don't know how i would reassemble the lamp
Electron gun sounds way cooler tho
it doesnt 

Modern hardware issue. Everything is either glued or potted
need to make a house out of aluminium
solution: build the lamp yourself
all upholstery is aluminium
if breaks just weld
Like one of deez?
also lamps connect directly to 220v and i'm scared of something going wrong
Into what?
so it kills you obviously. anti theft mechanism
kill you machine
These are them cables
The LED itself is usually behind the (hopefully) isolated DC circuit. So you're likely safe
It no go anywhere
the body can have a little bit of direct current, as a treat
i still have to deal with the power wiring
the lamp cant be properly disassembled without disconnecting the power
Worst thing is... this was an old Sony radio
Like why it have big microwave transformer
I thought it read the lamp can rather than can't lmao
there really isn't much that can go wrong if you have some distance to the lamp when you turn it on 
even if you're close the worst thing that could happen is a tiny explosion and a few sparks and some smoke or something
just really make sure everything is off while still working on it
could be impedence
could be signal coupling
who knows maybe its part of the audio producing hardware

Yeah, the problem is that means chay will have to power off the circuit to the lamp
Which may or may not be on the same breaker as other stuff
No.. you see that connector with the white and black cables?
That's mains power
i suggest lamp should become decorative ornament
it is what it is
other stuff will have to turn off too for a moment until the lamp is disconnected
?
its a regular lamp that you can just screw in and out
light bulb
Oh
having a lamp that you cant disconnect would be a problem to say the least 
oh, it's a blub
then yeah I'd still switch it off first but the assembly is probably more of an issue 
Nah, it exists. I just assumed lol
who needs light switches when your lamp has WiFi 
if you turn up the wifi power enough the lamp wont need a mains supply

Prisoner of War 
pal
Here the missing recrifying circuit tho
homophone of pow
This is why I never speak in english
yeah, if it's not a bulb then you just have to replace the entire thing
it's not too bad though since it's just mounting the new lamp and connecting two wires
just requires turning off the breaker
in fact if you turn up the power enough anything will become a lamp
have you seen this https://arxiv.org/pdf/2301.00250
nyope
detecting humans and their poses using wifi
induction lamp 
wifi
It does exists
also exists, kinda
You know, this makes me wonder if it can be used for selective jamming
Throw in a phased array antenna and you're golden
i dont want to be a lamp
it's just directional antennas 
main problem is distance 2.4ghz radio waves travel in air
just for reference
m
Throw more power at it 
i could be convinced
"let's overcome this nonlinear scaling problem by linearly scaling"
real programmer
It always works 
hello discord i do not want orbs
orbs?
your orbs are waiting
you could probably use some fancy modulation if you didnt care about wifi standards
1.2ghz would probably also work

aworb
The orbs will come and force you to do its bidding instead
it means Wi-Fi + LoRa
what does LoRa expand to
Long Range 
A brits saying this is always funny to me lol
LoRa (from "long range", sometimes abbreviated as "LR") is a physical proprietary radio communication technique. It is based on spread spectrum modulation techniques derived from chirp spread spectrum (CSS) technology. It was developed by Cycleo, a company of Grenoble, France, and patented in 2014. In March 2012, Cycleo was acquired by the US co...
proprietary 
Osmocomm
Send over the pipes
okay i see the issue
i am compressible in a hydraulic press
Finally, the ability to no longer have my home network skill issue at the exchange of being less powerful
jeff
osu!catch
next they will add a mod for std that allows controlling the cursor with arrow keys
ctb goes 2d
c2db
I know what to inscribe on my brick when I meet someone from this chat
c2b
unrelated or related
for marketing purposes, ctb 2

a sequel to the highly acclaimed and widely popular osu! gamemode
swift is... pleasant i guess
It is a good language
Catch the brick 2 mhm
now with physical components
maybe my opinion will change once i've had enough time with it, for now i'm unimpressed
i almost forgot i have to reproduce a hblang bug
the bug is that 2 * 16 is equal to 2247096596955168
ye i just dont fuck around when it comes to instakill electrics
Lmao
I know I already asked that but how do y'all find specific training data for fine tuning for LLM projects
Just google it?
you could i guess
there are some places with data gathered for llm training purposes
or just gather it yourself
ok but I need a specific kind of data
okay
Fine tuning implies you have specific goals that are not satisfied by off the shelf model. So I assumed it is indeed will be pretty specific and thus hard to find depending on the niche
Your best bet is just making one yourself
What?
No, as in, make the training data yourself
to make a llm clone of yourself you can for example download your discord data and then all your chat history data will be there to use as training data
I want to make Evil's idea of her using Axidraw
but instead I will have an LLM making turtle graphics programs
Then you will need to find a bunch of turtle graphics program
That, I do not know. As I said, if the niche is very niche, you may even need to make it yourself (the dataset)



plain js sounds like gigachad









