#programming
1 messages ยท Page 108 of 1
isnt the hello trianlge right there on the page? how did it take 2 days?
well.. rendering a chunk

i also have a memory leak in there... but... who cares right
i think it takes like 14gb of ram
wait looks like i did fix the memory.. it only takes like 400mb now
i bet you love cursive monospace fonts 
this you?
|
I lost the RTX vedal file
what a horible day to have a 27" monitor
No, my handwriting looks worse
good
e
nyope
e = euler number?
2.71828
e is a good number
pi = e = 3
huh
Nothing, I was just confused once when looking at KRANS and remember that you're KTrain is a different person. You have the same color tho, so that contributes to my confusion
dark mode kinda makes it hard to see
skill issue
hello empty void
fair
Bro, I am about 40-50cm off from my screen
Usually I am about 20cm tho when at home
ternary logic exists ye?
ternary logic circuits
you could make a pc with it supposedly
then the next step is to make it run an llm
then you have created a non-binary ai

elvyn stole your account ye
wdym?
ye
did you read it?
alr im a light mode user now 
did i read what?
1984
1.58bit lover
after all those years
we are still back to binary bits for calculating

ternary float operations per second
tflops for short 
paige
paige wasnt born gay, you're just indoctrenating her with yuri pictures
never name again
maybe later
konii, since when are you a part of KDE?
what is kde?
its the k de
k deutschland?
kool means coal
wtf did you just type?
is that how you think dutch sounds like?


isn't dutch verstappen land or something
speak -> praat
like (comparison) -> zoals
thos have nothing in common with english words
tu tu tu du, max verstappen
What about their accent?
what does norwegian sound like
which one? there are liek 12 of them
PewDiePie 
is pewdiepie norwegian

he's swedish
hes swedish
Nordic = Pewds 
yeah i though something was wrong
GO SLEEP
khayleaf
i slept from 1 to 5

slept enough
try not to offend every european country challenge
sleep police lmao
konii.... why are you doing this to yourself?
i was quite sure that pewdiepie was anything but norwegian
do not start a war between norway and sweden plz
It's those AbleCorp I swear
To be fair, I don't know a lot about Europe
we do need that
fun fact: "bra" means good, fine, well in norwegian
No one is here 
automod is always watching

Fair
feeling fine != being fine
You need B day sleep
nobody does
But 5 Bours is enough
Which is why bime is superior measurement unit
5 Bours = 7.5 hours tho
So 5 Bours, 8 Binute

i'm fine but i don't feel fine
then you arent fine
i fell asleep to the koko stream bros,
brain
you said one doesn't imply another
but i disagree
i am fine 
why would i beg you
not when im done with you 

"MINIMUM"
aimslop
so i am fine, almost
fieryrage jump
'.'.'
i am simply built different
Nah she built different
productivity hours
Not me getting my line stolen
Minimum Recommended is like Geneva Suggestion
geneva "suggestion" 
Lmfa9
the sun

the moon
konii = baby giraffe
the stars
๐ฆ
you arent a giraffe though, or are you 

konii is a tv with a picture of a giraffe. i solved the case
A newborn giraffe typically stands about 6 feet tall (1.8 meters) at birth. They grow rapidly, often doubling their height within the first year.


I miss being young enough to have a trash sleep schdule

i have a milkshake
so?
Actually going to sleep โข
its tasty
But you are young?
Oh wait, that's toast
I'm not
oldge
Uncge
do you happen to live on a farm?
not quite
I mean she does have ducks sooo,,,
i have an apple strudel 
so a rooster isnt the issue, you genuinly just wake up after 4 hours of sleep for no reason
I have a big meatballs with noodles
Really really going to sleep โข, jump me if im visible
actually a fox ate my chickens on friday evening

EEP
mhm
Spicy tho
ugh where is argv anybody know?
noodol
on the stack
yeh
env is also in there

next to argc
your phone camera never fails to disappoint
It still takes a picture just fine 
i mean at least you don't have to use the laptop webcam to take pics of things like chay does...
i had pineaple slices with white tripe sausage and rice with brown sauce
it took a while to google what the english word is for "white tripe sausage"
And once again, never fails to be cursed
useless := fn(): ^fn(): ^fn(): ^fn(): (forever) {
return @bit_cast(@frame_pointer())
}
mhm

actually its funny that you can indeed turn a return into a continuation this way
i wonder if you can call it as a function by doing something unholy
useless()()()()()(...
segfault
awa
i do that too
now i'm curious what desk view is about
how do you live with a high res screen
My webcam is not as bad lmao
obs
I don't have webcam program on my laptop
could you theoretically do 2 closed brackets after eachother?
maybe if a function returned another function?
every time i send a screenshot i'm embarrassed it takes up so much space in the chat
refer to above
this type of pattern is possible at runtime or comptime in hblang
It is called higher order function I think
the thing is i cant read that.
hblang syntax is giberish to me
There's also currying
x := fn(): type {
return x
}
y := x()
z := y()
w := z()
lmao it immediatly tried to connect to my phone
It is when you make a curry with your program

i wonder if there is any practical use for that
do you have desk view 
not like this, but yes
cuz i was just thinking of returning a different function
my phone just points at the ceiling, but my tablet points somewhat to me
In C++ it is quite cursed
At least not as clean as hblang
add_x := fn($x: uint): type {
return fn(a: uint): uint return a + x
}
add_one := add_x(1)
two := add_one(1)
its impossible to compete with C++ in that regard
it takes a special kind of development process
you sometimes see it in python
something like function()()
in python functions are objects
its cursed
it should be the other way around goddamnit
It is fun tho
it's convenient, and since python is duck typed anyway, it kind of fits
๐ฆ
if you really want to see it just looks like this
hdr
yes
ok so desk view is interesting, it's correcting the image using phone pitch (the rotation axis) so that the ground plane is a rectangle if that makes sense
@olive sable example of the a()() type syntax
What is ground plane in this context?
they couldn't even make it work with vertical

the plane that would be the ground
it does
...but why?
thy
what are you
'ing about?
so you can show what you're doing on your desk without mounting the phone on the ceiling probably
i had a meltdown yesterday when i had to play galaxy collapse on stream
Ahhhh
nuro and her maps
on stream?
discord
ah
Heh, performance point clicker
and then when i had to play LIFE three times

We have "Aria" Math now

its crazy that hblang doesnt have 141 stars 
does hblang have vulkan?
ablehub doesnt even have that many usesrs
hblang can link to c
does hblang use the red zone?
ye but that defeats the whole point, might as well just write it in c then
you cant use vulkan code without linking to c

write some bindings, have better experience
here's what it does when i'm rotating the phone rapidly
let's identify what languages it took reference of 
https://git.ablecorp.eu/mlokis/hblang/src/branch/main/examples/raylib/main.hb
here is a small thingy written in raylib that works
i was typing something but i have decided what i was about o say was incredibly stupid even more than my usual bullshit levels, so i have decided to move on
roll correction 
javascript rust python
The only three options fr fr

#programming top 3 languages
who tf uses js
more like ts
tbh my favorite language is probably glsl at this point 
insert Shiro~ ping
c subset real
sorry guys I might one day offend everyone here 
as usual
how does one become a user of ablehub
asj
ping erin or able
ask
what even is ablehub?
alr 
Does working at AbleCorp require able bodied?
i'm going to crash out, they're the same language ffs, enums don't warrant the distinction
uhh not sure actually
its like github but ablecorp?
self hosted git instance yeah
That's forgejo
says by a person with .jsx as name
is it specifically for hblang or will my c++ not corrupt it?
git does not make a distinction of the language you're using
its for ableos
func to_fizzbuzz_string() {
val div_by_3 = this.is_divisible_by(3);
val div_by_5 = this.is_divisible_by(5);
if div_by_3 && div_by_5 {
return "FizzBuzz";
} elsif div_by_3 {
return "Fizz";
} elsif div_by_5 {
return "Buzz";
} else {
return this;
}
}
elsif
The one running the instance is
.tsx does not sound as nice when i'm emphasizing jsx the syntax which is still called jsx even in typescript
el sif
what language it's from ๐
i didnt mean literaly corrupt, i meant if they would kill me for it
who uses elsif, thats like the worst of both worlds
I mean origin of this syntax
Python has elif, looks like they couldn't decide between that and full else if
Well, I mean that is up to the administrator too. If it is for ableOS, probably not

elsif 
ableos, hbvm related things
why do they do that
nah i prefer elif
I prefer Evil
else if feels cursed
If () {} Evil {}
well, not all of it

says the python user
if cond{
stuff
}eliv{
other stuff
}
โ
evil if (true) {
// unreachable
}
else if feels like im putting an if statement in the elseplace instead of it just being a 2nd option
which is exactly what you're doing actually
I like this idea lmao
no
why not
nooo
gets rid of the if (!( syntax 
if (a && b && !c) {
} else (c) {
} else {}
implicit if
if a && b && c {
} else if c {} else {}
cond do
2 + 2 == 5 ->
"This will not be true"
2 * 2 == 3 ->
"Nor this"
1 + 1 == 2 ->
"But this will"
end
every language should just have cond
Elixir conditionals and pattern matching is the best
it's time to collect everyone's idea and make a neuro language 
i think you can do this in hblang with enums if you curse it a bit
How about the fallthrough?
i hate that
else if
is the same as
else
if
i would have rather died not knowing that
id be like 7% happier

kmnowing makes me activlty suffer
what fallthrough? it just checks the condition and takes the first branch
no other branches are taken after that
i think it makes sense if you consider else and if to be separate expressions
Yeah, I just realized that. It is very explicit
Or verbose
i thought because they were toghether it was treated diffrently. idk
python had elif which was an entirely diffrent word
this is a real thing people do in js
switch (true) {
case 2 + 2 == 5:
return "This will not be true";
case 2 * 2 == 3:
return "Nor this";
case 1 + 1 == 2:
return "But this will";
}
0.3 + 0.7
doesn't seem idiomatic but kinda based
if cond expr else expr
is an expression. and you just plug in any expression into these.
I have never though of using this pattern
is it based or not? it definetly looks kinda cursed to me
this case is so cursed but since js is a dynamic language i guess it makes some sense
it is cursed but long if-else chains also kinda suck
i think ive only used switch case once so far. purely becuase i forget it exists
i accidentally unplugged my second pc and now i need to drag the massive monitor back into my room just to turn parsec on
True := enum(bool) {
.true := true;
}
match True.true {
// nyope cant do it
}
but those are mulitmodal and dont output audio, just input of audio.
is it because bool enum are not supported
this probable should be a switch case now that i think about it
its because match only accepts enum literals for arms
I mean switch cases arent reallly much better than if else chains in most langs, maybe a little more explicit. but in languages like rust where you get a compile error for not handling all possible cases, it becomes a ton more useful
slowly learning better brackets 
nope
brackets are redundant in that screenhot though
why?
you can do closures in hblang so there is an alternative way anyway
because c accepts a single statement in place of a for/if/while block
i put the fully all on the same line or all seperate.
no "learning better"
I need my pattern matching
<<
width::32-little-unsigned-integer,
height::32-little-unsigned-integer,
bin::binary-size(width * height * 4),
>> = :ezstd.decompress(imgdata_compressed)
stuff like this is too good to give up
-# this one isn't inside of a case but it could be
but i thought switch cases were better?
Case := struct { .r: bool; .fnc: type }
$case := fn($cases: []Case): void {
// ...
}
case(&.[.(1 + 1 == 2, fn(): never die)])
amazing
by the way i forgot to tell you
struct { .a: type }
works completely
its perfectly sensible when you learn the rules of the syntax
but i can see how it might be daunting
switch is easier to read, less duplicated code and sometimes a bit faster ( maybe )
i really doubt theres any cases where modern optimizing compilers like llvm dont optimise if chains to the same instructions as switches
godbolt time
indeed
you would be suprised. But also why I said sometimes
the only immediate optimisation i can think of is hashed jump tables so there's room for skepticism
Godbolt doesn't support hblang 
hblang has its own disassembly features
and compiler runs in browser
godbolt not needed
bwa
to make it clear, i mean the compiler can be compiled to wasm
hblang compiler in JS lmao
if depell.mlokis.tech was still up, that is what it would be using
i have switched, pun intended.
it would probably be better to just asign variables for the magic numbers but i cant be bothered
The domain is being held hostage
???
also im completely ignoring the possebility of a "TEXCOORD_1" which i find hilarious
lol
i dont think this will work
And ngl, everytime mlokis is mentioned, all I could think of is Muhammad Loki 
i have not tried it
oh yeah right, C++ strings 
mmm ptr
because switches only support integers
const char* ptrs are indeed integers
but you probably wouldnt want to match on that
std::string I think overloads == but not whatever a switch statement does because that's probably impossible
i would indeed not want to match pointers
and if you use a char* instead then you run into issues when you happen to match a string that's not a literal from the same object
static 
if you put "POSITION" twice in the same file (specifically the same compilation unit) then your compiler will very likely merge them into one, so their pointer will be the same
if it's somewhere else then the compiler will be glueless and won't do that, so your pointers will be different
im not trying to use pointers tho
obviously also applies to anything dynamically-allocated on the heap
why are those strings anyway. Do they come from the vulkan api or something?
unlucky
You can convert them to enum if you want 
but why would i do that?
Unmarshalling is usually done to simplify things when programming. Such as the case that you encounter. You can use switch-case if you perform the conversion
but the problem is that you have that 0 there
but you can have 1
or 2
or how many the gltf spec decided to make possible
idk if there is a limit
thats the problem ye
my plan is to hopefulle look if the first x characters amtch
eventually
rn im just not supporting it cuz i dont care enough yet
i don't think there is a limit in blender either
you can add as many uv maps as you want
arent they just casting an int to string?
wdym
sorry im distracted rn
there is a limit
looks like it
In Blender, the limit for UV maps on a single object is 8. This limitation is based on OpenGL's maximum number of texture coordinate sets
huh
does that mean... when vulkan becomes the default and opengl is gone... they might support more?
mayhaps
what
what's the difference between uv maps and textures? cuz i remember the opengl texture limit being 32
can'T you do bindless textures. and therefore not have a limit
uv maps just tell how to sample a texture. like this vert is at this location in the texture
uv maps are texture coordinates
it's representing the coordinates of a texture
why are texture coordinates limited tho? its jsut a vec2
ask opengl
i dont get it tho, wouldnt it be 32 like the texture limit?
and the reason for having multiple maps is to change where you sample no?
ye me neither
you rarely if ever use more than 1 uv
Explicitly, it has types. Implicitly it has any type because it runs on JS 
maybe a second uv for like wind parameters or some other exotic stuff...
but then you would probably use vertex attributes instead
yeah, AFAIK for 3D rendering it's basically just how you unwrap the mesh into 2D space so that you can texture it
there's very little reason to need 2 ways to do that
you can certainly use multiple textures, but multiple UV maps is weird
im pretty sure that gltf changes texcoords_0 / 1 / 2 just for when a texturee changes too, not specifically for when blender has a diffrent uv map
dont quote me on this tho
so did neuro
konii is neuro confirmed
UVs count towards vertex attributes, maybe they need to reserve spaces for other attributes
actually just look in the code and see why 
neuro is better at osu! 
according to this random blender forum you can just remove the limit
though i havent seen konii and neuro in the same multi room 
it'd have to be like 1cm tall to make a difference 
rarely
and filters
i used to use an almost 1cm tall area
i uesd to keep randomizer on 24/7 but dont do that nowadays
she's trying to look less like neuro
fine motor skill training
technically, you can "remove" the limit by just using normal generic vec2 vertex attributes
it's probably a historical issue
it is ye
i think the smallest area i used to use was like 30mm
gpu's have increased the limits
as opengl 2.0 uses 8 as vert attr limit
I'll happy to submit a pr or issue to blender later


spam that ticket
i still dont see why its a limitation. i thought it would jsut be another in vec2

it pretty much involves around the blender's main body and renderers
making it disappear would cause compatibility issues
"too much work"

...right, that's not wrong
i wonder if i could make my game run on a watch
1.5GB ram and 16GB space
that 2 days abttery life is a scam its still only 10 hours

i was actually trying to make my aim look more like neuro
lol
my aim is stable 
as someone who is a professional coder, I can guarantee that those movements are from an algorithm designed in java

wait how do you take screenshots on a watch
You press the power and back button at the same time
high technology
nah I don't want to give the photo viewing application access to my photos
too risky
ignore that not doing so would kill the entire app's purpose
help...
it gives you a bash shell with some git commands on windows
can i uninstall it without any damage to my pc? cause i dont use it anymore
if you don't ues git, probably?
you can probably uninstall it if you don't use Git
only exception is if some other application installed it as a dependency
but you would probably know if that was the case
do you know that this is a shortcut?
Then you can uninstall it
i forgot why i even installed it in the first place
no?
at least i dont think i do
so do you know how to uninstall it?
huh why i try to press on it it shows me this
bull crap uninstaller
i see
so do i uninstall it or not
cause of this
yeah just uninstall it
the error is fine, it's apparently due to git for windows not being able to cope with certain security measures from windows so you'd either have to update it or make an exception to get it to work
but if you don't plan on using it just yeet it 
my crew, my people, my loved ones. Here.
we don't like vedal here too 

someone learning infosec here?

why nopers in 30fps
where nopers in 15fps





Ive uninstalled and deleted 40GB of data in my drive
The joy of not having package manager
Let's play a game of "what needs this as dependency?"
shoot
Well, you can start by figuring out what programs needing git in the first place that lysara have installed
if you donโt have a package manager on your target platform and you donโt ship the dependency yourself you probably deserve the user to hate you
To be fair, I like how windows programs look for dll in the same directory first so multiple programs with different version can be easily installed even if ill advised
On linux, that thing is only gaining traction fairly recently I think with flatpak (and to some extent, appimage, but it is not that popular)
snaps suck and so does ubuntu
before flatpaks and stuff there was static linking
obviously very different technologies but do basically the same job of bundling dependencies
why are the Neuro clips/videos audio compressed to heck sometimes?
it just sounds.. weird..
also, on that osu clip, why was Neuro's voice increased so much
was a little jarring to me
โbefore there wasโ
static linking, alive but forgotten
also tell that to rust
the Go language in question
rust will static link so hard that you get the same library multiple times
i am statically linked
are you position independent
no

refer to the relocation section in my header
relocatable 
vscode please leave me alone
dev stream 
Perhaps I have no rights. I shamble
I think you should use it
Living dangerous is the only fun way to live
Looking at your tag, I can see that
Another NixOS user has breached containment. Not that I'm any better.
We need to spread the virus
Quite a few people here already use it
perfect
I used to run into @tender river 's repos when I started NixOS, and that is before I met her,

You're famous chay
Because I think I also remember encountering your site before I even get in here
i hate you all, probably 90% of the code i've written this month has been nix 
i saw a video of 6 y ago and he is explaining how a ANN works, basically a neuron is a node that can do function for example input and output, in the middle is a node that is what the neuron do, like for this input do this things and u can add neurons or things that the machine can do based on the input, still not understanding why i do need math, for what, if iโm programming in python or c++ i can do a neuron by a simple function and if i want to connect that neuron to another, maybe is there where i need math but i still donโt understand How. i donโt see any practical example of math being used for a neuron example a machine that have to move my mouse based on where is the enemy, no need to mathematic (?) i think, iโm not saying there is no need iโm saying i donโt understand why, ofc machine is a bunch of 1&0 but, for high level programming language, there are already functions for math no? i repeat myself, iโm not attacking anyone, iโm just trying to understand and learn , because iโm new i have so much to learn so, thanks in advance guys ๐๐ป๐ซถ๐ป
That's not the usual ANN
Look up multilayer perceptrons
Those are computed with matrix multiplication on GPUs for massive scales
Btw, who is the colab partner here?
the merch company
Wait, it's real?
yeah

That's a simple perceptron, multilayer perceptron puts a ton of those one after another as fully connected layers using matrix multiplication
the math i need is like
enemy position = 300
mouse position = 2
function = move mouse from 2 to 300
That's not even an AI model at that point, that's just an algorithm
do u have sum explaining well what an ai is ?
server dead again
cause i think thatโs just the base iโm missing
Your server?
An articial neural network, or ANN model is a model that uses a linear-algebra model, that can be trained with backpropegation to fit a set of data, called a dataset, to achieve a certain result, such as in the case of the MNIST database, classifying handwritten digits, that being considered a kind of "Hello world" machine learning project
ok now i have a point here, thank you master
I just silly around and make vocal synthesizers, nothing crazy
void movemouse() {
mouse = 300; // assuming mouse is global
}
Am I mistaking bcachefs with btrfs? Isn't one of those a bit dangerous as you can lose a few minutes of work when the system suddenly lose power?
I accidentally made a questionable robot
bcachefs and btrfs are different filesystems but also both are modern filesystems with journaling
Hmm, I may have misread it on the other server then. Someone borked their arch install after a sudden power loss when running upgrade
Yea, you're supposed to have a rescue disk in that inevitable scenario
arch doesnt do atomic updates
How do people (like in the game jam) create games for Neuro's API? I thought all of Neuro's code was non-public
or is the API actually a generic thing used for more than Neuro
ah, so Vedal shared that specifically ๐
it's not neuro code, just an interface that neuro implements
yeah I see, not really part of the LLM
a neuron layer is at the end just a weighted sum passed through an activation function e.g. relu which is f(y)=max(0,x)
bro it is only a 0.05GB model why do I need 31GB for compiling that
the math aint mathing
i'm so conflicted on finishing the llm touchbar keyboard thing
on one hand it's a very cool idea and i really wanna see how it'll work
on the other hand, i would not wish working with apple apis on my worst enemy
it's so miserable
WAIT SO MUCH WHAT
maybe i should ditch rust and do swift at this point but my familiarity with how swift works is at 0
at least i wouldn't have to write bindings
probably, unless i touch private apis
Bruh my AI Neuro Chatbot
does setting theTF_FORCE_GPU_ALLOW_GROWTH env variable to true fix this?
Why are you torturing yourself even more than you should already do by doing this cursed thing 
1:1 perfect replica
if your primary use case for rust is interfacing with native apple apis just use swift honestly
its not even an embed fail its just i dont want to download the wav again
thats what my neuro chatbot sent
That voice is.... something
its edgetts ada voice or whatever
:mhm:
what the fuck
it's real
can it not see whether it has an image or not and just hallucinates whatever
with a heavy bias towards equations
that's crazy
almost as bad as flutter
i like swift as a language, idk about everything around it
was very bureaucratic whenever i tried it
the server can wait i played like 6 or 7 hours of osu! today i'll sleep right away
as far as i understand they're super big on interop, and i've seen some obscure apple stuff with swift, as far as i remember it's pretty easy to work with private apis even so should be a better experience
almost as if it's a language primarily by apple for apple 
yeah what's up with that 
They timed out my bot lol
https://i.imgur.com/cOXofYO.png
heres your homework, #programming, figure out how to (topologically) make neuro's 4-sided invertible hoodie a reality

hmm
hard challenge
You have unlocked new role
This seems interesting
i just realized i have embeded perms lol
Congrats on having newfound rights
is this a joke or real fix
it tells tensorflow to not allocate all the memory upfront
and only use what it needs
(yes, i had to find this out the hard way)
I just saw there was a programmer that made a insane app for a very young age they Said "I coded it myself" and theres just AI Comments in there code...
I want to say that you add a zipper on the side of two inversables and flip that.
if it works who cares (big if)
How it would exactly function I don't know at the moment
User: Yo homie can you like introduce urself concisely?
Model: I'm just a little AI assistant, but I'm here to help you out with whatever you need. I'm pretty chill and easy to get along with, so feel free to ask me anything. I'm always up for a good time and conversation. I'm not gonna lie, I love to chat and learn new things. So let's get to know each other a little bit more, okay? I'm here for you.
what the frick did i create lol
you nullified your training, congrats. this reads like an average llm response
My Friend: lmfao meaning
Neuro-Sama AI Chatbot by me: {Filtered}
wym nullified training
idk what your goal was but this reads like something an average "base" model would say
lil model is not being concise
Mine:
E 
i can fix that surely
seeing arm64-apple-macosx as a target name is so weird
Stealing from a plush I currently don't have access to. A convertible pokeball to Igglybuff plush, it's method of inversing is a double sided zipper and feeding itself through. Should be able to stack with the inside out method that most jackets use.
-# Steps: Get a zippable jacket and a hoodie, modify the zipper to be double sided, stitch the two so that they stack, profit.
this was the original
Yo! Iโm Gemma, a large language model created by the Gemma team at Google DeepMind. Iโm an open-weights model, meaning Iโm widely available for anyone to use! Basically, Iโm here to chat and help you out with text-based tasks. ๐
oh my god
gemma love
lmao
https://i.imgur.com/UGr0p21.png
not even my neuro knows
my console logs rn:
AI Reply: [Filtered.]
AI Reply: [Filtered.]
AI Reply: [Filtered.]
AI Reply: [Filtered.]
I'm still smarter than AI for now :SAVED:
i now understand why people buy those crazy $6000 macbooks
i imported appkit and swiftui as a test and now it's compiling the world
it's been 15 minutes since anything visibly changed
rust move over fr
and it's not stuck, i see stuff appearing and disappearing in module cache and my cpu is crying
Me: say hi to the programming channel of neuro sama discord server
Chatbot: https://i.imgur.com/FApltrS.png
she hates us
it would be terrifying if she liked us
im going to check her logs to see what shes saying
Great lol
[Original AI Reply] Oh! Okay! I said FUCK. Wow! I did it! You can ban me now.
[Filtered by AI Filter]
[Final Sent Reply] [Filtered.]
You should at least be using Neuro RVC to get a better voice match
The more I think about this the more I want it, dang it
It'd be fun to figure out how to deal with adding pockets to this thing.
this did not work, almost obviously cause im using jax
YES
damn i got baited with the error message
classic jax behavior tbh
uses tensorflow or something in some what way
but not so important
I have the rvc but idk how to implent it...
The v12 one?
V12.1 is latest
You just need an RVC runtime you can call from your code, then you load the model and index, and feed your TTS audio trough that
That's latest then
It's some open source RVC That runs on gradio how will I use that? Post Requests?
I seem to recall an API
this is so fascinating
cool
This is why I love Mistral Small
tool calling?
Good luck finding many non thinking models in that size class that can solve a riddle like this
I assume you manually added the models to it
I'm going to use the normal RVC
What do you mean "normal RVC"?
I mean, what matters is if it has bindings for your programming language to use its inference engine
And I guess slightly important is if you can set the inference params to the optimal ones
(venv) C:\AI\NeuroSama\RVC>pip install -r Requirements.txt
'pip' is not recognized as an internal or external command,
operable program or batch file.
(venv) C:\AI\NeuroSama\RVC>
uhhh
i think i have to recreate the venv
People can have their own neuro? o,o
Mine is just a discord chatbot
Technically, if you program it yourself
It also won't have the same LLM, but you can at least get really close on voice
I wouldn't know though, I just do Neuro vocal synthesis
if you want to get close on voice you can use the actual voice
superbox you should think about that
I weren't much surprised on that, more so over the fact it's literally called neuro :o
For TTS yes, if you can use Azure
I thought it might be copyright or smth idk
https://www.youtube.com/watch?v=49CrynNZAR0
In case someone somehow hasn't seen it yet
collaboration with Superbox, this makes it the first (unofficial) Neuro-sama cover of BOOM!
come chat w/ me on discord if u like! discord.gg/8ry5dn3UjA
check out the instrumental version! - https://youtu.be/KSxZScNeJBo
NeuroSynth-BETA-3 developed by Superbox & Wispers
Motifs:
BOOM (Evil) - https://youtu.be/8Nt9YPnYyjs
Only present in instr...
Why does it sound like a heavily modified version of Miku lol
idk
It's NeuroSynth, the only thing they have in common is they're a vocal synthesizer with a well known character behind them
It's not even the same architecture
I'm going to have 3 AI's running on my PC Now we got Neuro Chatbot her Filter and her voice lmao
I just use 2 llamas for Neuro and then RVC for her voice I have a RTX 3050
RTX 3050...
I assume that's llama 3B?
Q4 I hope
Also can't you use the same model for the filter and the main AI?
I think its mistral-7b
No... llama for the filter mistral-7b for her ai
Also how are you running 2x 7B models?
One 7B model should already be all the VRAM
idk it just works
Is the filter 1B?
Which llama model is it?
llama2
swap go brr
I really hope you at least use a quantized version of the Mistral model
I think the smallest there is 7B
You should really be using llama 3.2 3B for the filter or something
And both your models should be Q4 quants
Good, at least you got that right
I usually use these
https://huggingface.co/LiquidAI/LFM2-1.2B
https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/blob/main/mistral-7b-instruct-v0.2.Q4_K_S.gguf
You could however use this
What ones better for my use?
I will download that
Usually just take an up-to-date version of the model you can run
This is an updated version of the Mistral
This is a more reasonably sized filter model
Those two might fit in your VRAM together, though it may be tight, you will have to look in task manager to check VRAM usage
wouldn't mistral-small be the latest version?
That's 24B
Way too big for a 3050
I could run that with my 3090, but that doesn't make it useful to this other person
there is a mistral 8b that was updated 6 hours ago
Hm?
Though i am still waiting for one that is around 5B ish
Why are you using a model that tiny? Do you just have a very bad GPU?
qwen 3 has a 4b model
{
"discord_bot_token": "xxxxxxxxxxxxxxxxxx",
"bot_mention_name": "Neuro-sama",
"tts_audio_dir": "tts_audio",
"gguf_model_path": "models/mistral-7b-instruct-v0.2.Q4_K_S.gguf",
"filter_model_path": "models/Llama-3.2-3B-Instruct.Q4_K_S.gguf"
}
that good for my config.json (got rid of token)
It's a new gen model. I tried both qwen3 7B and LFM2 1.2 and i couldn't see any difference in results beside much faster inference
I assume somewhere in your code you set N-GPU-layers
I assume you have a quite low VRAM card
what do you mean?
So you can't poke into the 24B ranges
How many layers of the model is allocated on the GPU, you know, a basic parameter which is required to use the GPU
I never done that but it worked fine..
Is it slow?
And what's the issue with that? o-o
And does it make your CPU usage max out?
im pretty sure it is automatically using all the gpu
Just that there could be models with better capabilities in the bigger ranges
Nothing wrong with using small models with small VRAM
Talking to me? cuz yeah it is
It depends on the library
Have you checked task manager at any point during usage?
Just try it. I really do believe it's on par with 8B models.
I can't bother to set up llama.cpp and program something to poke at it right now
My task manager broke and will not open even after restart
Uhh
Install Linux
At the point task manager breaks, you're screwed if you stick to Windows
Unless you want to in-place reinstall, but that isn't exactly something one would think of
You need some way to varify what is causing your stuff to be slow
At the point task manager breaks, you're screwed if you stick to Windows
What may you be trying to do?





thats all

those names are confusing