#programming
1 messages ยท Page 247 of 1
Maybe it'll help, who knows
It will give more of the data to the model at least
Also increase batch size as high as possible
maybe 
The original code had a Batch size of 20
The more batch size the better, it improves compute efficiency/utilization and the averageness of the model
Well if you have the memory, make it way higher
Remember, NeuroSynth is trained on 128 batch size and it still overfits
I have only 8GB of VRAM
so I can't really go that high
Well adjust it based on usage
if you are using max ram then you can only change these settings
Only load one batch of images into memory at once
Oh, I set it this time to 5 and as you could think it didn't go great
Now trying 1
Luckily I need to generate the samples only once after... Yeah, Batch size 1 isn't gonna help
Batch 1 is not gonna go anywhere
You need to figure out what in the code makes it load the entire dataset at once
And change the loader such that is only loads one batch at a time
4K images with samples per images at 1 which gives me a small chance didn't help either...
If you were only loading one batch of images at once you could have infinite images
x_train = (np.load('x_data.npy').astype(np.float16) / 255.0).astype(np.float16)
y_train = (np.load('y_data.npy').astype(np.float16) / 255.0).astype(np.float16)```
If I knew how to do that 
Yeah that's gonna limit you
You gotta replace that with a dynamic loader
I am at my limits of even understanding the code 
Then get better
You gotta know what you're working with to do good in machine learning
It helps quite a lot for me to be able to adjust NeuroSynth's trainer code too
CodeParade really knows what needed to be done for a good model
Well that part they just did that way because they clearly have infinite VRAM
CodeParade said I should stream the training data in while it is getting trained.
Yeah that should be what is there
The model was made 2018
Or late 2017
Streaming the data is indeed what you want
And I have no idea how to do that 
Well you need to learn
It's a wonder rn that I make coherent sentences rn
The data loading code is suspect one
Controversial idea: Going to chat
oh god oh no
Blender I see
blender is going to make it so you can bundle the lines together in shaders (geo nodes, etc...)
don't get me wrong, incredible idea
flashbacks to minecraft ME systems though

no channel limit tho 
I kinda hope I can complete a lot of new NeuroSynth/JP data next week
sjfkldfllk;j
That's crazy
i need to do something else but i want to model now
ill have to look where i put the old blend file 
I can't even model stuff at all
True, it has it's upsides
RAM Modules?
blender like eating ram
superbox singing arc
jok
where's my stupid screenshot
Surely not as much as Firefox
also I keep putting of learning to sing in any capacity and it annoys me greatly
Surely you can just make a vocal synthesizer of your voice

or I could just learn to sing
firefox has never used the full 64gb before on my pc so im pretty sure blender uses more
Firefox uses more
voice train to neuro and learn to sing 
I have confirmed cases of 23GB+ usage
close tabs
Surely
I already unloaded most of them, didn't help much
:3
with firefox you kinda know when you do something that takes a lot of ram, cuz you have 200 tabs open
in blender you misclick and oops you have 64Gb of ram fully loaded, with a page file of 40Gb next to it
well I was looking for something else but
bro has a 64K x 64K texture in there 
idk what I even did to cause that
I have only 64GB of DDR4 
"only"
Yes, that is what I mean
64gb is a lot
Yes that is very little
16gb is bare minimum, 32gb is medium, 64gb is large
64GB bare minimum, 128GB decent, 256GB big
Yeah, but I have another PC that has in Total 64GB too
om
I would say 32GB is ok
64 bare minimum for some ai stuff, sure
but uh, most people will never use 64
you say that, while also complaining about being broke 
Now were talking serious memory 
Now that's kinda getting there
"kinda"
Well if I wasn't broke I would have 256GB DDR5
Anyway
64 bare minimum?? what the fuck is bro on 
im perfectly fine with 64gb, i know i will literally never use 512gb
Two web apps
but sam
Web apps really like to leak memory
they don't
i have a leak so i must buy more buckets ah
jsut fix your fuckign memory leak bro
literally the only instance i know of is 7tv
Discord: leaks memory
YouTube: leaks memory
Twitch: leaks memory
AlieExpress: leaks memory
Google: leaks memory
GitHub: leaks memory
bruh, why they all so leaky?
noone will ever need more than 128kb of memory

I have confirmed them consuming multiple GBs of RAM each after a few days of running
how is it possible i have better memory safety than multi-billion coreperations?
Heck yeah:
noone will ever have more than 1.44MB of storage
Now I increase the Batch size
Oh did you manage streaming data?
they are all electron
maybe thats the reason (well or atleast react)

Probably
Superbox, I'm sorry that I didn't do it
But VRAM usage is better: VRAM: 25.2 MB / 2471.3 MB |
2471 MB ????
Oh yeah that's way more like it
Now make the batch size 512 and you're on your way
I have the Asus NVIDIA GeForce RTX 3070 Ti TUF GAMING OC EDITION
there's no way a 3070ti has 2GB of vram
That is the usage btw
Left is "Current" and right is "Peak"
def print_vram_usage():
if not gpu_var.physical_devices:
print("No GPU found")
return
for i, gpu in enumerate(gpu_var.physical_devices):
details = tf.config.experimental.get_memory_info(f"GPU:{i}")
used_mb = details['current'] / 1024**2
total_mb = details['peak'] / 1024**2
print(f"VRAM: {used_mb:.1f} MB / {total_mb:.1f} MB", end=' | ')
print()
return```
Should separate by | instead of /
i have 30+ tabs of github permaopen in a ff based browser, no it doesn't leak, same with youtube. vanilla twitch just sucks perf wise but doesn't leak. can't say anything about aliexpress. google is a funny entry no the fuck it doesn't leak anything it's literally a website billions of people use daily including me no it doesn't. as i said literally the only leaks i encountered in my browser is 7tv and discord
Well, Should I yeet the model I have rn that I'm training and start from scratch?
Well it was more likely Google drive since it was just a google.com domain using a ton of RAM
google drive does seem like the type of app to leak
I don't know how I commonly see YouTube tabs and stuff using upwards of 1GB though

at daily usage i have 7/64 gb, in blender i have maybe 20 gb used. its only really when i fuck up and misclick that i use the full 64gb while blender tries its best to not shit itself
App leak?
the type of app to leak memory
If I knew why my memory gets sinked so fast over time with Firefox often using 18GB+ I would fix it
iirc, the game-jam game has some memory leaks i never bothered to fix
its on the scale of only a couple megabytes so i didnt consider it worth my type
My current mem usage is 41GB
That I didn't know, I'm sorry
Total full system

what are you even doing to be using 40GB?
System ram?
holy
Jesus filtered
System Ram?
that uh
I don't know, most of it is usually Firefox
I'm suprised the nas is handling it that well
Like half of it is almost definitely Firefox
brother what are you doing with your firefox????????????
That is faster than my DL speed 
That was my question too
Geez
firefox only uses 1GB on my hyprland
wait lmao what
I think someone merged a bunch of video files together and somehow went from 50mbit to 500mbit
Interestingly in at least system monitor I can't even account for all my system's memory usage by adding up the major numbers
chrome is great with resources
It can be great
Though system monitor sucks, btop just doesn't seem to have sort by mem usage
Hiiii konii 

holy fire this is hdr too
no it just objectively is when there's firefox skewing the statistics
so what you mean to tell me is that you repeatedly lose your ram and because of that you say 64GB is the bare minimum?
or at least 10 bit?
god awful browser
Well I would very much like to know where it goes
And I personally am too lazy to switch browsers.

firefox has a whole page dedicated to breaking down memory usage doesn't it?
webgl? 
if not two pages
I use Google Chrome since 2019 at least
@sage crag 
about:memory yeah
sure if you have javascript bindings
bruh
oooh
Oh now I get why you are confused mlntcandy, Sorry.
Hem
Doesn't really line up with what system monitor tells me
System monitor must be wrong
cant import c libs, can import wasm modules though
so if you have a wasm module for webgl it would work fine
I'm sorry mlntcandy 
blink is good
Sorry if this is completely off topic but does anyone know of
a website/resources that -
Regularly updates a timeline/list of influential AI/LLM research papers based on the number of citations made off of that paper
If it doesn't exist someone should do that
Id assume it would basically be a web scraping program that keeps a list of papers checked and then moves through all the papers published so far
And just like counts citations for each paper as time goes forward
since the AI Boom
Would be a valuable resource for reading up on how we got here and what the mainstream stuff
You can even probably code it so that papers that were later disproven are labels such
Would probably make more sense to make it into a kind of visual branching tree diagram where the most referenced papers have brighter connections each other
Idk if I'm explaining this properly
No I'm not begging btw just a thought I had
Did you forget that I'm German?
So yeaaahhhh this thing kinda uses a bunch of memory and I would like to know why
๐ฆ
my brain hurts what is going on
Where is the memory even going???
I'm sorry mlntcandy
The breakdown doesn't look that useful, there's no gigantic outliers this time
Eh don't want to leak anything by accident
superbox, firefox literally has a built-in task manager to see where your ram usage is
I use:
And yes, I'm still not chaging the theme
Sure
im so mad
i add feature thats supposed to prune bad moves
=> it prunes good moves too
If that one's only 3GB thogh
- where is the 15GB+ coming from
- WHERE IS MY RAM GOING
ye i was about to say
Still, it's not 15GB total for sure
sort by memory
That is some false positives for ya
it would be easy if it pruned best moves only
no
it literally adds random prunes and nothing else
sometimes its good sometimes its bad
0 consistency
only prune when its non-pv non-check non capture no nothing?
0 difference
Hi vituha
I don't know, doesn't look like 15GB

How's going? 
ok some random previous version beats 1500 elo stockfish
veru good
i discovered that that stockfish that was DESTROYING my bots wasnt actually 1350 elo as it logged
it was going max power
it says "elo limit 1350"
but the box in settings isnt checked and its ignored
Oh do the small things seriously add up to that much?
So I guess then Firefox is indeed using 15GB
yes
comback from -5 on 1700.. stockfish just loses on time basically.
it wins but spends more time. ends up having 5 seconds when my bot has 30
you have 3 + 2 + 1 + 2 x 0.7 + 4 x 0.5 + 6 x 0.3 + 5 x 0.2 + 8 x 0.1
so 3+2+1+1.4+2+2+1+0.8
roughly 14
the small ones add up
as white stockfish finds mate too fast and i lose
i have to figure out whats the deal with these phantom upgrades im making or else this will be my cap
I'm running out of tabs to unload and this thing still uses so much RAM
This is all that's left
oh wow thanks @nocturne olive
Wha
swapped to torch and instantly recognized my gpu lol (i mean after setting up cuda and stuff)
Well yeah, Tensorflow is known for having removed GPU support on Windows
removed? or just not up to date?
i think i made it work before, i just had to downgrade shitton of stuffs to make it owork
Yeah it existed on early versions, but not there on modern any more
ah alr
๐ thanks
why would some courses in coursera recommended tensorflow then as the starting point for neural netowkr?
May be a bit outdated
And I'm outdated
hi
Hi, Welcome at #programming
TIL!
This just sent me on a sidequest, turns out you can go to about:about in FF to get a list of its about: pages
:o
So is there on Chrome:
broooo pb why did you annihilate the instrumentals so much in the cover 

So
So
so what you mean is that the entire app doesnt work?
oh is the entire platform just dead?
the browsing features works
videos work on my desktop still
the video i was watching before still works at least
ok ye other videos dont work
good thing i was watching a 40 minute documentary
videos don't load and this now shows up
the link leads directly to this
"have a problem? blame adblock" -youtube private documents 2025
oh it's back for me
brother even yt music is broken
i got a 1 month free trial on tuesday cuz they patched the cracked one i was using nd i need to find a new one
bwa
spoke too soon yep dead again
i hasve my entire playlist downloaded
Seems about right
your automatics have been purged
i don't think i have 2 months of music if i combine all of my music like all of my playlists from every streaming service i ever used, all music from itunes era and cd era. 2 months is fucking crazy
probably like 90+% of the people on twitter uses yt so its understandable
thats what you get with terabyte i guess
unrelated, but please photoshop. what the fuck are you doing? why doe the image have gray holes?
ipods' selling point was that you could carry your whole library because they had hard drives
and they were like 50gb
100gb
That'd still work for me
100gb is not too much for like movies, but for music its a lot
that's my point
i mean i didn't say it so i expliicitly confirm
not all of them could play video
the later ones could no?
i wasnt a breathing creature back then so i have no clue how the different gens worked, but i seem to remember ipod color or something
or am i confusing it with the gameboy?
Yep they did, I remember having to juggle storage on that when I had it.
far too long ago
5th gen and later could play video apparently
idk what 5th gen means exactly but sure
thatโs like approximately when you were born
07 was 6th gen i guess
i thought they skipped a couple but apparently they were annual
wait
what the fuck is that release schedule
ok so they did skip one after the 5th
01 -> 1st gen
02 -> 2nd gen
03 -> 3rd gen
04 -> 4th gen
04 -> ipod photo?
05 -> 5th gen
07 -> 6th gen
dead
Man the wikipedia is a mess with all of the skews
there's a unofficial 7th gen that was just a 6th gen but they made it slimmer cuz of hard drives getting smaller, but they also added a new "genius mixes" feature so its basically a new device kinda
Which one of yous are buying an iPod touch in 2020
no it was a proper new gen, better hardware and all
Draw your line, when should it have stopped
on the 6th obviously
7th gen made no sense
the fucking iphone 11 came out the same year
everyone was on streaming services in like 2017 already
on their phone
iPod Touch 7th Gen iPhone (around 2016-2019)
Wi-Fi only Cellular (LTE) and Wi-Fi
Apple A10 chip, but underclocked Apple A10 chip (faster clock speed)
so its jsut a worse version of the iphone
YT works fine for me, but there are def a lot of people that it does not
probably a botched rollout update
I created a web app using Firebase Studio: Doctoral Suitability Assessment Scale A scientific, non-diagnostic self-assessment tool to evaluate your readiness for a PhD program. https://dsascale.vercel.app/
I'd consider 6th gen also a bust due to in line with iPhone 6
yes, it is essentially an iphone, but no biometric id, no vibration/haptics, no earpiece speaker, no cellular modem
but it was 199 so actually not too bad, compared to the iphone being 450
You can view source code here: https://github.com/ZhanZiyuan/doctoral-suitability-assessment
the worst models, not the full upgrades versions

yes but you already have the fucking phone
do i?
in 2019 you even had apple music for android already
find one reason to buy the ipod touch in 2019
there are literally 0
ipod touch is essentialy just a cheap phone, i would have mained it
I tried making an apple account recently and it didn't let me verify so I can't use it
you couldnโt for the reasons i mentioned above as differences
who cares about biometric if it saves me 250 bucks
i mean if your requirements from a phone are only facetime on wifi then sure
you couldโve
the only actually deabreaker one might be the cellular
i hav no clue how that works tho
no cellular means also no 4g?
No cell service does suck for a "phone"
yeah
A plan to back up the capital?
to me the word cellular is a nothingburger buzzword with almost no meaning behind it. i read it as "phone go bzzzz"
also this is fucking crazy but facetime is insanely backwards compatible, i still can facetime from my ipod touch 4th gen which came out in 2010 and the latest software is ios 6
it still works
i mean it did like a year ago
thats to be expected no?
no doubt it still does
why

how is it expected
Netflix was also pretty backwards compatible, I think 3ish years ago they removed their android 2 support
itโs a proprietary video calling thing
idk how facetime works
why would it work on a device 15 years old
my frame of reference is facebook messenger
your face gets thrown to the other person in realtime
fun fact, they wanted to make facetime an open spec
initially
while there are no capital cities in Japan.
but a patent troll ruined it
same with imessage iirc
or maybe not this iโm not sure
does tokyo not count?
Yes it is but not legally.
huh
It was somewhere else, not sure of the exacts
lets see how long youtube lasts
something went though at least

is it even like allowed to sell a phone without cellular in the eu
I guess you probably can but noone does it
why wouldnt it be allowed? 
i would expect that with nowadays mobile platform tech, especially from facebook (i know netflix isn't facebook i just thought of them) because they're known for stuff like server defined ui and created react native which is js which means js bundle patches on the fly "circumventing" the app stores, so they're much closer to web apps in that sense
idk something in my head about emergency calls
because you're supposed to be able to do them from any phone
it is nearly 2am so I'm probably not thinking perfectly straight ๐ญ
fair
I see https://www.appluslaboratories.com/global/en/news/european-e112-mobile-emergency-call-regulation-(eu)-2019320)-to-come-into-force-on-march-17th- but haven't read any of the specifics enough to make a valid opinon on it yet
wifi only phone is certainly a choice
o wait this is for the ipod
idk makes more sense because it was literally an mp3 player that could download new songs, but when they were popular downloading over mobile data would've been expensive and slow ig?
apple literally wouldnโt let you download apps >300mb through data until like 2018 or something
๐ญ
that wasnโt a setting or anything, literally hardcoded
my numbers may be off but there was that limitation
the limit was 200mb actually
it was 150mb until 2019
wtf
is there still a limit?
no
i donโt think so at least considering how many people have unlimited
and i donโt think iโve ever seen it
after telling chatgpt to tell me I'm correct it says im correct
Conclusion:
For a phone to be legally sold in the EU without the ability to connect to cellular networks, it must still be able to make emergency calls in some wayโwhether through cellular access or another form of emergency communication [...]

i havenโt looked too much into it iโm probably wrong
in order to sell a phone without the ability to connect to cellular networks [...] make emergency calls by connecting to cellular networks
thanks chatgpt
but i couldnโt find anything in effect
fair
(36) It is important that users should be able to call the single European emergency number "112", and any other national emergency telephone numbers, free of charge, from any telephone, including public pay telephones, without the use of any means of payment. [...]
hmm
I think it depends if its classed as a telephone or not?
"we never called it a phone there's a reason why it's called a shitbrick 3000"
well like I think you can still sell tablets just fine?
idk I haven't looked at them in ages
from a glance samsung sells https://www.samsung.com/uk/tablets/galaxy-tab-a/galaxy-tab-a11-grey-64gb-wi-fi-sm-x130nzaaeub/ which doesn't advertise cellular
so I guess its probably allowed?
what is this godawful font
timeline:
- 10mb: initial limit with the release of app store in 2008
- 50mb: no exact date, apparently slowly increased it
- 100mb in 2013
- 150mb in 2017
- 200mb in 2019, end of may
- literally one week later in an update it now shows a confirmation window instead of stopping you, beginning of june 2019
i was on android back then but i remember that vividly because peers tried to download games and couldnโt
hi guys what do we think of this
๐
casual user moment
i don't even use linux
shame
I would
but asahi linux seems to be the only real option according to the small amount of research i've done
and they don't yet have proper support for M3
nor M4
Windows good for software, the OS can go into hellfire
only good for software because people don't develop for Linux
...because the userbase is so small
Correction: software is good for windows
*as a consumer
I'm just kidding
??? (same email btw)
Part of it could be Twitch compression
But also the mixes on karaoke streams are sometimes overly vocal-oriented
I personally mix deep to make the vocal more balanced with the instrumental
hope nobody clicked that
There was a scam link somewhere?
Some shady trading link or smth
Speaking of karaoke though, why is there none of the new stuff in the drive from yesterday?
it used to be where compressing the mp3 from the google drive with even 96kbps ffmpeg aac would sound less butchered than the stream, yeah
not sure if that's the case anymore
Yeah stream audio is horrible somehow
I have no idea why
Could be that Vedal's system has to do some mixing on the fly to play both the instrumental and vocal but it kinda breaks
I blame Unity to be honest
yay it works now
anyone else getting this?
expired certificate prob
bit weird for it to expire already
you can check the SSL Cert
oh so it did expire
no wait I misread that
there you go
it was supposed to expire 2 minutes ago tomorrow (according to sent back cert)
can someone else check the cert fingerprints for me?
I feel like this is way too short for production
what's the fingerprint?
fun
I'm seeing this cert too
(it's mismatched to what I got, which I think means something is interfering?)
oh something must be interfering on your end
check your network security i guess
i'm on train wifi rn so I don't expect that the train wifi is playing nice
What site is it 
isnt 3 months kind of short?
what else has a .app domain?
wait, you're on a public wifi?

yup, though I don't expect someone to be trying to mitm
maybe vpn will help 
just to be safe i would do that too
its pretty weird for it to say expired when everyone else has an updated cert
oh fuck is my vpn even enabled on my vps
doesn't happen for vedal.ai
so I just need another .app domain that's obscure but is an actual site to see if network admins block .app domains with smth weird
i cant load youtube :LULE:
no it's bc of public wifi 99% sure
definitely a train wifi issue, ye
hmm, guess it can be automated, yeah
Accidentally stumbled on what looks like a really good deal for some excellent headphones (under half of its intended price): https://aliexpress.com/item/1005008888888308.html
(Beyerdynamic DT900 Pro X)
Though if I scroll down below on that page, there seems like a lot of pro/audiophile grade headphones also for much less, huh, never knew they can be gotten that much cheaper over there
I heard aliexpress can be risky 
Risky as in
You buy something and something much worse quality arrives
That's not the risk
That's the expectation
Aliexpress is cheap because straight from factory and has no QC
The risk is that it never arrives because the shipping is bottom of the barrel
I haven't bought much from AliExpress but I see they have some form of customer protection, like refunds in case of no product arrival, does anyone have first-hand experience with them?
in order to sell a phone without the ability to connect to cellular networks [...] make emergency calls by connecting to cellular networks
thanks chatgpt
for some reasons plugin, that i decided to try a while ago, just not want to work. clangd installed and config just like readme.md, but just not work Warn: No Lsp servers found for "<project>/coroutine_reference.cpp"
what would you recommend for lsp client? just common one, but there a lot and idk what to choice, not want spend 3 hours just to read them all and extra 6 hours to choice
# vim
I know
got his 7GHz+ god binned 13900 engineering sample from there
wait, you're on a public wifi?
Hi, there. I created a web app, please feel free to try it: Doctoral Suitability Assessment Scale https://dsascale.vercel.app/
The source code is here: https://github.com/ZhanZiyuan/doctoral-suitability-assessment
wait 4.5 haiku is actually good?
What is that?
Oh useless garbage
ding ding ding
Why do i feel like some messages have been repeated 
had the same conversation yesterday
or the day before 
what is this based on, like where are the questions from, how are they correlates to doctoral suitability 
my graphics card is AMD
Then everything should work OOB
OK. Thanks for your feedback. I'll try to solve these in the next version.
Now it's a prototype by GPT-5 and Firebase Studio.
The intro page of the website provides some references.
aliexpress is best.
just takes quite longer to ship than amazon in general
but its basically chinese amazon, just as reliable
and its recommended to only buy from a seller with a lot of sold items and high ratings
Yo
Yo
I need dev help
What happened
Making self talking function
const MEMORY_LIMIT = 729828;
const MAX_REPLY_WORDS = 82200;
const TEMPERATURE = 0.8;
Morning
function tokenize(text) {
return text
.toLowerCase()
.replace(/[^a-z0-9'\s]/g, "")
.split(/\s+/)
.filter(Boolean);
}
Good morning
then you should be good
To be honest Nvidia works perfectly fine too
I run a dual GPU setup and have had no issues at all
โ
there's just an extra step to follow on fedora
So vibrant
new record
who thought that would be a good bitrate for 1920x1080@25fps 4:2:2 10bit
also seemingly with interlacing artifacts for good luck I guess

I've started messing around with Python recently. A lot of stuff still goes over my head but I managed to write a Discord bot connected to an OpenAI API. Which has been fucking hilarious so far, but I can already tell it has a lot of limitations regarding pretty much everything besides standard plain responses with a modified personality.
So I was thinking of letting my next hobby project be training a language model on some data and see how that goes but ngl I'm worried about computing power. I have an Geforce RTX3060 and 32GB of RAM on my current PC which I really only use for work (Design) and I have no idea what this involves in terms of how beefy a PC I'd need. Like I said, hobby project, I'm not expecting it to solve math equations or rationalize philophical questions I just wanna play with it.
but still, Arch vs Fedora good battle?
arch is good if you want very high customisation
i recommend fedora kde for a decent mix of customisation and sane defaults
they just completely different approaches
arch is generally bleeding edge and rolling-release
fedora is cutting edge and a mix between different versions and rolling updates
neither would necessarily win, but i'd recommend the average person, gamer, or programmer, to use fedora kde
arch takes a lot more setup time with a lot of things
archinstall mitigates that a bit but i wouldn't trust a sometimes-faulty python script to install my os
i personally use fedora workstation, the gnome flavor of fedora, cuz i'm just used to it 
personally:
"I use arch btw" said every guy who uses arch
Arch causes traffic congestion 
Ye ur good compute wise
Thank you for the opinion, hopefully it'll get useful to me if I ever get over my patience issues (I got really used to 1-2 day shipping from local shops)
nice documentation Microsoft 
i don't think my bot is any smart yet (play against stockfish) or i have something wrong ig
even i see that thats not good piece development xD
Femboy 
rook here looks like bot took first move in list
it can happen for many reasons
yeah im not looking too deep yet and every position probs had the same value (no extra value for positioning yet)
you 100% need positioning value
its much more important than it feels
from my experience these are actually basic things that need to be in any alphabeta-like heuristic bot for it to be competitive:
- quiescence search
- static evaluation with positioning
- alpha beta pruning
yeah but i really did not put in enough time so im still getting the basics to work
probs not gonna end up submitting, at this point im just in it to see if i get something halfway decent to work
actually good improvements that work and give significant profit:
- pvs
- move sorting
- transposition tables
For me move sorting made it like twice as slow
how do you sort it
I tried many sorting orderings
Well I ended up with a put captures first thing in the end
It mostly doesn't do too much and I hope it's better at scale

i saw people say using the chess videos form Sebastian Lague are a good start, so im working along that without looking at the code for now
I don't know if the reason it was so slow was because it was having to do a lot more memory copies per searched move
i didnt list iterative deepening because it makes bot x10 times slower
but it unlocks many other things you can use, and it becomes hard to tune to make it better
it doesnt make it that much slower but you start wasting most of your time per turn on last depth search which isnt even valid because you have to stop it in the middle to not lose on time
so in reality it looks like its x10 times slower
but you cant really use that time without iterative deepening anyways

alright i guess i'll wrap this alpha beta whatever i did into 1k tokens and then go scout my old bot a little maybe i can find a way to improve it with all my gained experience
sounds like a plan
It always looks like a dev blog. I wish it was a forum and not a channel so that we could track these discretely but at least people can ask for help or feedback and get it relatively fast
We have #1336777692070023271 but it has to strictly be neuro related, nothing for general programming
I don't think a lot of the regulars would vibe with it if it was a forum
at the very least I don't think I would
Honestly, same here. Took a bit of time to think on this, but it'd just feel more isolating than just a single chat. Sure some of it is blogs, but also a lot of memes and just a place to chat.
Why not both then? Programming general for memes, help and one-off feedback; and a programming projects forum with regularly updating threads
It will declutter the discussion also
Because that'd be an extra channel to check, and honestly. A new chatter came in the past and tried to propose it to mods; they got denied because they aren't interested in expending the dev community
Oh well, it would be brilliant but I suppose #programming was more of a fun addition than substance for Vedal then
He'd only want the neuro related stuff to have stuff because he can use them. Anything else not even I could see a reason he'd want them.
Just to host a community is enough of a reason. This chat is essentially Stack Overflow without the toxic forum culture rn, and it does nothing but good for all those who just want to build a small app or a niche proof of concept without any intentions for a big release.
So it's not a completely alien idea to expand on it, but if we lack mod resources then nothing can be done about it. It's better to have no platform than an unregulated platform; otherwise I may just log onto an imageboard like Kiwi Farms or Sharty this very instance and try โasking for help thereโ.
I honestly see nothing good out of forums, they'd still be isolated and if you were trying to connect them it'd look no different from self promotion. If you want people to be aware of your project, of course you link to it.
The issue that I have is how fast that scales, and at what point it just becomes a soulless message board with a new purpose of promoting projects, and fragmenting what few people we have here.
Ic how it could turn this way. It's a very valid concern.
I, on the other hand, am mostly looking at another successful forum (#1336777692070023271 ) and just applying the idea of how nice and cosy it would look like.
Either could be true, but it's irrelevant if no-one moderates it. It will become a spamming platform before a self-promotion one without any mods patrolling it.
I'm not one to say how often mods are here, I have my opinions but it ultimately isn't a point I'm using for argument. But what I will say is that you'll need the consent of the chatters here otherwise it will fail, and I have a feeling not many of them agree with your stance.
I'm of the point that it is nice and cozy as is, and anything more to spread out could easily hurt that vibe.
I Agree to disagree
Ok so we just booted up an 8xH100 box from e.g. Lambda GPU Cloud. This is costing us about ~$24/hr, so there is no time to lose. Environment setup Clone the project: git clone git@github.com:karpat...
Less is More: Recursive Reasoning with Tiny Networks
https://arxiv.org/html/2510.04871v1
Reminds me of this
#neurotic-neurons message holy shit, that's so good
it is a tad noisy
#programming message this is mine by comparison, but I didn't do the fancy stuff they did
My PC just hard-reset
I would like to know why
Just what you get when you have the instrumental on-hand
I wonder if you could potentially get better results by making a song an instrumental (that doesn't have one) first, though
maybe too much loss
You'd need to reproduce the exact same settings with the exact same instrumentals and plugins to get good results
The way that technique works is essentially you apply a reversed copy of the instrumental, and adding the waves together negates them
Reversed as in vertical
Not horizonally reversed
ah, yeah, it'd need to be a true instrumental to work better than just a pure vocal extraction
only thing phase worth doing without a true instrumental is a phase fixer with two models, but I haven't had luck with that
and that's to reduce noise, so you can use a fuller model and then denoise it and can potentially get a better result
Need UI help, this is my game's UI screen rn right, when you hover a tower in the tower shop panel on the left where should the description pop up? To the right of the panel? Below it? And how big should the description panel be? 
desc would be smth like
Arrow Tower
500G
Fast firing, medium range single target damage.
Hp: 150
Damage: 20
Aspd: 0.4
Range: 6
Or should i just have it be relative to where the mouse is at while hovering etc
hdd
What I meant, words are hard
tl;dr if I read from it wrong it then it goes to 100% active time
Uhhhhh
it's on an external mount atm
What does "read from it wrong" mean here 
no idea
Like certain files?
it's a mystery to me too
no idea
I think so?
it's either specific files or reading too fast
If you haven't checked for bad sectors yet you might wanna do that 
how does one
Uhhhh
Actually idk if windows's disk management thingy does it but there's definitely third party software, surely one of them is free
I use aomei's partition assistant for all this stuff on windows but it's paid so idk
damn chkdsk makes it very unhappy too
joy
I'll probably see if I can convince Linux to be happier with it later
and by later I mean tomorrow probably
ntfs3 my beloved
also there is a dubious ssd being backed up
somehow only reading at 33MB/s
on video files that are several gigabytes big
Checked Crystaldiskinfo yet?
Unless it doesn't have any smart info I guess
Actually
If it's backing up the SSD to the HDD that would kind of check out speed wise
Obviously 33MB/s for a hard drive is still terrible
nah
the ssd is unrelated
I'm also trying to back up both
ssd is hilariously slow
considering I was hitting gigabit writes to the backup share earlier

idk why but slow drives always make windows chug for me
surely nothing is wrong
chkdsk has given up on giving me an estimate
Slow drive is slow nowaying
most hilariously my laptop has been sacrificed because a job I thought would take 4 hours max is taking 15
"in the end"
wishful thinking assuming there is an end
it's such a pain that this stops me from doing a disk dump
something somewhere in the chain is trying to be helpful
windows is just especially bad with slow drives
feels like every interaction requires loading 50mb
Assuming those are different drives and sda is unmounted should be fine
Though you might want to increase the block size unless you want to wait 3 business days
read 3 gigs then died
I/o error my beloved
time to do a bad sector check
or well let smart do it's thing
I finally caught it btw
92 power on hours glueless
wait no
I am illiterate
6000 hours
whoops
allegedly my SSD has 28000 hours
wait no that's not my personal one that's the society one thank god
Interesting that it claims there have been uncorrectable sectors at some point but then it also says no sectors have been moved


The price of nitro basic has gone up to 36 per year, it was 30 not even a week ago
Morning
Not horizonally reversed
damn
Morning
?
Hru
Gud
yes sir lol
make a horizontally spinning vedal smh
anyways gn Sam I'll cya tmr
Though you might want to increase the block size unless you want to wait 3 business days
The block size of what?
ntfs3 my beloved
smh
?
The real part of this is not bothering to add music to it
Whoever runs @graceful pasture and @paper creek should at least make up their own messages 
i have blender open on my main monitor
O my god
Bro came with army
opening obs was already more than enough effort
Hmmm
@open copper I think you might be interested idk
Hi I need to run Voxtral mini ai model, for transcribing live audio to text. I'm coding in rust, and I was just interested how dificult would it be for me to launch the model directly from rust, instead of using python?
I'm having dificulties finding material on the topic of running .safetensors files directly without a use of some lib. that supports running the model directly
The problem is that safetensor files on their own are kind of useless
They only store the tensor data, you still need a mapping of that to an actual model
Ahh, ya I would assume I could reimpliment it myself, I'm just not sure how much efford it will take.
This is the model: https://huggingface.co/mistralai/Voxtral-Mini-3B-2507
I HATE MATRIX
I wouldn't bother tbh, you'd need to write an entire inference platform
I WON'T DO AGAIN IN 4D
Lol, now I'm interested whats the backstory?
Alright noted thanks
I was testing 4d
X y and z what after 4d demension
w
Matrix cracked
, graphic out of control
(X and y and z = one side โฌ๏ธโ๏ธโก๏ธโ๏ธ โฌ๏ธ)
What do you need 4d matrices for anyway

Are you asking chat what's the fourth axis?
Traditionally, it's XYZ for three-dimensional spaces and UVW for other concepts. So you could have XYZU for a purely abstract projection.
If you're a proponent of a time-space physical simulation, then time is the fourth axis. It's traditionally represented with T.
Otherwise, whatever software you use may already have a variable pre-defined
W to represent the 4th dimension is more common
What I thought but I don't have enough information to prove they 4d exists
4 d start out of graph maybe thinks
Totally unrelated to the names, are the pfps of the same person but diffrent angles? 
I don't think so, different hair colour
Drive dead or disconnected
That happened on my 8TB drive once and it was a loose cable connection
I just hit limits
On what?
Equation is soo hard it feel like unsolveable
inbefore it is unsolvable
Is there a way to resolve a domain name to the ip address of another domain name? (for example, have x.com resolved as whatever the current ip address of xcancel.com is?)
Can't you just add an entry in the hosts file like x.com xcancel.com?
I thought the question was about how to find x.com through xcancel.com
In a shader, the 4th one is depth
To see that its actually a mirror
Hosts file only does IP <-> domain
Doesn't that require the original ip as well? (which means you have to change it every so often)
Your best bet is something that auto redirects in the browser
There are extensions for this
any recommendations for SVS project with finetune instructions avaivable? as far as i can see espnet is the best looking one
yes, not much training data though unless i would be able to somehow synthesize it from a normal speech samples
Well there is RVC to SVS distillation
Though known to not be that amazing
This is done using RVC -> SVS distillation
Though I'm also working on making a real proper dataset, labelling takes ages
Do note that you are expected to have quite a bit of GPU power available
But yeah this arch is pretty good
Don't try to compete with NeuroSynthProject though, Neuro and Evil are my SVS
Well they're mine now anyway
What is that reaction supposed to mean?
jk, i want it for shitposting
I assume you're trying of some random voice you found an already formatted SVS dataset for
no, i was kinda hoping there's a way to finetune some base model to the target voice with a self-made small dataset. Unless no and that's not how the typical ai cover shitposts with modified lyrics are being made?
Ohhhhhhhhh you want RVC
You want this tech
It's trained using just voice data, no need for labels
The input is an existing vocal
THAT is what those very badly made AI covers use
I see what you're looking for
That's so easy to do
i thought RVC can only convert the voice, but not change the lyrics
Ohhh
You want to change lyrics
Well that's not as easy but you can still do it with RVC + an existing synthesizer
You could achieve the base with UTAU, or a machine learning alternative for higher quality
Then RVC on top
That is how this was made
So you want this right?
i'm looking for something like this cover https://www.youtube.com/watch?v=P_x5aY2-_FM
Didn't embed
Oh Drive
That's done by the person singing with a very terrible RVC model on top
It's the author singing their own base
damn thats neat. Sorry for the XY rant
Yeah it's pretty silly
And it makes for a very horribly bad voice output
It would definitely have been better with my model but unfortunately they didn't use it
But yeah that's how Drive was done
Now you know
If you'd like help training models, let me know
I do quite a bit of that


does anyone have a vmware or kvm/qemu hardner so that the secure browser variations cant detect vm
stupid test service wouldnt accept list.append() as a valid python method
Here is my information
( I'm useless guy who wants gta 5 on my pc but my pc is on has ended even hhd is damaged
(2gb ram
500 gb hhd
I3 )
Sadly I really working on phone to code and working on Google colab to run code and save in local file )
If the ps3 can run gta5 (badly) on 256mb, then you should be able to do it mildly well on 2gb
On low settings
1080p 30fps
maybe
Hello everyone and welcome back to another Budget Builds Episode where today we're taking a look at GTA V, but this time we're taking things a step further and finding out...Can you make it run with just 256MB of VRAM in 2025. Turns out after 10 years on the market GTA is a little bit too unhappy running with such a small amount, but with a lot ...
that video is about vram but still an interesting video
actually on integrated graphics sharing that 2gb ram you'll probably need to do 720p or maybe even less
2 gb ram my windows is hogging 6 on its own
just dont use windows then
im pretty sure windows uses less if there's less available tho, so in reality it hsoudl be fine-ish
Yo !
is it newly installed 11 or with revios/atlas at least
Just updated to 11 yeah
Does it consume more the more ram i have
heard revios people say it does
Windows 10 and also 11 do in fact consume more RAM if more is available
I don't recommend using Windows these days
yee it does
I should make a "unused memory is wasted memory" sign to tap
"if those kids could read they'd be very upset"
we love using 64Gb of ram just for the sake of it 
but ye i do agree
caching is based
Extremely fascinating paper on using an LLM to build Expert Systems via generating Prolog facts: https://arxiv.org/abs/2507.13550v1
programmersss
i wanted to share one of my favorite features in gaming
procedural animation
i wish more games utilized it more
I'm looking at running my own LM for a hobby project really. I'm kind of torn between using a pre existing base model or training one entirely from scratch, I don't know much about programming in general it's mostly a hobby thing I've picked up but I found some interesting projects on training LM's from scratch. I think where I'm torn is that it's kind of funny when an AI isn't super intelligent. That's what I'm partially annoyed by with the OpenAI API







