#programming
1 messages · Page 321 of 1
ye example
New lol
me put this on same line 
Aww 🤩🥰
because
it just one operation
swap
neural 
nny?

Aggressive inlining I see 
me compiler
I am now for neuro
It did hurt me
Neuro did ask..
sorry i cant understand you
try machine translation if english isnt your first language
Me sandwich and saan.. I'd love the songs and the beasts

As a sandman
I'd love to bee
.
Neuros beef.. mand her
Shi love her soo much
It's fine with what you do
Neuro is so cute
yes that is definitely neuro
also this is #programming….
I know
skynet 💀
I c
-# haha get it because C sounds like "see" haha- okay I'll get out :c
HOLY
LATENCY
♭
this was supposed to send way earlier :c
neur


got any pixels?
here at sam.dev we have a wide assortment of different pixels for sale
for only 5 neuros per pixel
⬛ 🟦 🟫 🟩 🟧 🟪 🟥 ⬜ 🟨
thats brown?
i thought it was red
I'll take 2
no refund? 
this is a pyramid scheme
no refunds
🔺
any scheme i do is a pyramid scheme 
its my branding
not the main branch. and the useful stuff branch is jsut what im currently workign on in the engine
but im 3Dmodelling a lot lately so even that isnt updated
🔺 stuff has passed mostly, now its jsut my brand
what if i have 1 ssd for linux, and 1 ssd for windows, but i partition the windows drive to have windows on a separate partition on the drive?
cuz i know people say to put linux and windows on separate drives, but that should be fine no?
why not just VM the windows
when i will get my gpro x080 and new mobo i will try linux
mint or zorin are the most user friendly ones right
cuz i have work to do on windows
photoshop and visual studio
i want that on something i can trust to be mostly stable, as stable as i can expect form windows at least
windows should be installed first
then linux
iirc
well if on separate drives
then yeah it doesnt really matter
Shouln't matter yeah
idrk
Or you can run WSL direcltly in windows
only if you want a specific distro you need to tinker
cos the default is a guiless ubuntu
i have bought a 990pro to be in my mobos main slot,a nd a sn7100 for the secondary slot.
linux oes on the main slot, and i want to contain windows in as litle space as possible in the secondary drive while using the rest of that drive for gewneral storage
nah. i main linux
then just VM the windows. Unless any of your win apps really hate being run in a VM
the whole point is separating windows from linux
i don't think a windows vm can use gpu accel? i do need that desperatly
it doesn’t matter even on one drive. only thing it will affect is if the bootloader setup process picks up windows automatically and if you install linux first you may want to rerun that so it picks up the windows bootloader and adds an entry for it. or add it manually
there shouldn’t even be issues with the partition layout
people say that windows and linux on one drive cause problems but in my 8 ish years of dualbooting desktop linux (windows 10 already a thing) is windows sometimes, very rarely nukes the efi boot record for your linux bootloader if it’s in the same efi partition as the windows one. annoying but not exactly the end of the world
so in theory it would be faster to and possible to put both windows and linux on the main ssd?
i have always done it like that
so in theory i allocate like 30gb for windows, and 30gb for linux. then i can make both use the other partitions for storage?
Don't

why does it need to be on the same disk anyway
ideally yeah
many reasons
like you know
a laptop
for example
it doenst, but the main drive has dram so it would be faster
990 pro 2tb main, sn7100 2tb secondary
look i’m not suggesting anything but i have admittedly too many drives and i still have linux and windows on the same 2tb drive
i will never be using the a full terabyte for windows let alone 2. so i wanted the space available for other things
make the windows partition like 300gb, that should be plenty enough

if it’s not your main system that is
this is my main system
i mean windows
oh
operating system
ye im mainly on arch
If you're worried it's not enough leave some empty space after the windows partition
i just need windows for Photoshop, visual studio, and potentially for VR since my linux is still glitchy
but steam games should be able to get installed on different partitions so that should be good
visual studio takes up a fuck ton tho … maybe 300gb is too optimistic
There is an easy solution here though 
with the 10 toolchains and sdks you’ll need
it cant actually use 300gb? 
i thought it only used like 80gb with my sdks and toolkits and such
not 300 no
"only" 80
compared to 300, yes only 80
ikr 
i don’t think it is.. microsoft has like 10 competing technologies to qt
the whole qt stack is like 40gb 
you should allocate at least 250 GB to your system partition, with 500 GB being the recommended "comfortable" sweet spot.
bwaa
why windows so big
a real-world installation with updates, system restore points, and hibernation files typically occupies 80–100 GB over time.
surely i can nuke hibernation
i was planning on tiny 11 i guess
myeah 512 is the minimum I would use for a Windows boot drive
Yeah you don't want hiberfile on a small drive
but depending on what you do you might want to keep XXGB for pagefile
Especially with your 64(?)G of ram
ye
My work computer has 128 GB of RAM and a 512 GB boot drive. Took me second to figure out where the space went. 
surely i can disable hibernation 
You can indeed rip it out
One thing I miss from windows is how fast the system goes into hibernation

It takes like a minute on linux (on my rather crowded pc)
I used to use hibernation on my PC just to prevent my mouse from waking it up. After fixing that I just use normal sleep 
My pc doesn't post half the time after waking it so it's kind of a requirement for me 
Support my channel on Patreon: https://patreon.com/MegaLag
Check out:
Jelte for Honey Data Drop: https://x.com/j3lte
DataRequest's full investigation: https://www.datarequests.org/blog/honey-data-collection/
Install Cookie Guard: https://chromewebstore.google.com/detail/cookie-guard/ifjhcahbhkfojdmkndpkmkffbjnefido
Chapters:
00:00 - PayPal's ...
yeah, I watched it already
can't believe another video finally came
1 year after is crazy



Would anyone here be able to recommend whether 2160p is worth it over 1440p at 27"?
it is more than doubl the amount of pixels
in terms of gaming higher refresh-rate is a lot more noticeable. but for productivity applications you can fit a lot more text on the same screen real-estate
a lot sharper too
yeah, tho I see people saying that it's not that much of a noticeable upgrade unless you go with larger screen
but you also have other people saying that is noticeable
i think on 27 inch, 1440p you can still vaguely see the pixels, but 4K is already starting to reach the point where you cant
depends on how far you sit tho
1440p is fine for most people, but there is a noticeable difference between that and 2160p
I guess kinda close, so 4k might be better. Either way both will be an upgrade over current 1080
personally though, id not consider it my first priority
i think it might be a good idea to go to a store and see the difference for yourself
my first priority for gaming would be refresh rate, and then if its OLED or not
ye
OLED is so much better for motion clarity than LCD
you can get lcds that get close, but oled is near perfect
response time is technically separate to refresh rate, but you wont notice a difference
absolutely true, but at the same time eh, no proper electronics shop nearby
oled's contrast is a godsend too
the ability to turn off pixels means you can actually use the display at night
without being blinded
sometimes lcd displays will have multiple dimming zones but it usually sucks compared to oled even then
yeah, oled looks great tho I barely game so idk if the insane price jump over isp is worth
if its not for gaming then 4K is a bit more worth it
more stuff can fit on the screen
if you dont care about clarity and colour depth, then lcd is usually fine, but i'd say i'd rather have a good 1080p oled over a 2160p lcd, having used both
mostly depends on your preference unfortunately
idk about 1080p oled, i think 1080p is jsut not worth it for me generally.
kinda want an oled, tho I don't need one any time soon and I'm already used to isp. And if I'd buy one I would probably go for a 4k one. But at this point the price jump is like 2x-3x from original isp that I was considering, so better for a future consideration
tho yeah, thanks for input, helped in making the choice
lmao
at least it's not super critical, like, accidentally putting in links that direct to a 404 3 times in a row

i just had my biggest aura gain
by pulling out my briefcase pc in front of the customs security
security was pretty understanding too

mf even joked i could launch a missile with it
this made me think of a certain Bigroom song
SV2 Teto is certainly different from any Teto I've ever heard
I pray looser RAM timings make my PC stable
the slop thickens
@tender river 

honeycomb
warburtons fruit loaf with cinnamon and raisin 
christmas is after halloween and before the new year

that does check out
except that all 3 of those are exact dates
so not every day inbetween is christmas



the shit that i do at hotels
drink water, eat candy 
That's my average hotel getup
please recommend a cheap and interesting game
what sorta stuff do you usually play?
5D chess with multiverse time travel
What is it?5D Chess With Multiverse Time Travel the first ever chess variant with spatial, temporal, and parallel dimensions. It's the first ever chess variant with multiverse time travel!
Features
- Sharpen your tactics by solving a collection of multiverse chess puzzles.
- Practice against four different AI personalities, each with different stre…
$5.87
7721
uh-uh... um-mm... I usually play... I don't know
here's a screenshot of my games
beyond my intellect
ftl maybe? it's on sale
War crime sim
hmm... thank you
that's why, WHY? there's not a single message, but as soon as I write without pressing "reply," someone is SURE to write at that very moment😭
is this the same mobile game from an ad?😂
FTL is now available in 9 languages! English, German, Spanish, French, Italian, Polish, Brazilian Portuguese, Russian and Simplified Mandarin!
************************
The free expansion, FTL: Advanced Edition, is available now! Content additions include: new alien race, events, weapons, playable ships, drones, and mor…
$2.49
61054
84
I want an L4 and a blackwell pro 6000
As the resident California boi, checks out
I'm glad I'm not living in that area 
The system that Texas (and some other places) have been implementing over the last couple of years has helped a lot with grid stability
wtf is this
It works pretty well actually
i cant access google
Date/time wrong or DNS
wrong time is unlikely I think, that should be a different error
youtube works
outdated DNS is possible but would be weird
what does it mean
i cant even google
oh can you give me google ip
so i type it directly
use bing (real)
If you use 8.8.8.8 as dns and it's out of date there is a problem
ye, VPN provider DNS may do that
or a local cache (like the built-in one from Windows, or something like systemd-resolved or dnsmasq on Linux)
holy viber
yeah okay reconnecting to vpn helped
I feel like people don't understand that what Neuro is asking him to do isn't as simple as it sounds... Especially since it's a question that's mostly been avoided by not developing something like her
What are we talking about here?
local DNS cache
can store an outdated A/AAAA record, causing an HTTP(S) request to go to the wrong host
Yann is just plain incorrect here, he’s confusing general intelligence with universal intelligence.
︀︀
︀︀Brains are the most exquisite and complex phenomena we know of in the universe (so far), and they are in fact extremely general.
︀︀
︀︀Obviously one can’t circumvent the no free lunch theorem so in a practical and finite system there always has to be some degree of specialisation around the target distribution that is being learnt.
︀︀
︀︀But the point about generality is that in theory, in the Turing Machine sense, the architecture of such a general system is capable of learning anything computable given enough time and memory (and data), and the human brain (and AI foundation models) are approximate Turing Machines.
︀︀
︀︀Finally, with regards to Yann's comments about chess players, it’s amazing that humans could have invented chess in the first place (and all the other aspects of modern civi…
Not gonna lie Hassabis has really been on a roll recently
this reads like pseudoscience
brains are the most exquisite and complex phenomena we know of in the universe (so far) [
], and they are in fact extremely general [
]
"☝️🤓"
who is that guy in the video
someone important?
Realistically I don think the answer is something that can really be properly quantified to xisting norms. Trying to do so is going to cause problems.
Meme or genuine question
the weather, a galaxy, quantum fields, microbiology ecosystems
Yann LeCun, very clever guy and former head of Facebooks AI efforts
Left before Llama 4 to start his own lab

there are lots of things that are extremely complex in the universe 
to be honest the idea that brains are the most complex, even that we know of, is like saying the earth is the centre of the universe

Its pop scienece since its worded for a general audience but not pseudoscience
pop science may as well be pseudoscience for all the good it does
Demis Hassabis is a ridiculously clever guy and the brain is his area
You don't have to gatekeep a point from the audience you're presenting to for it to be salient
This is already 3 minutes long and i haven't even made a proper buildup yet... this is gonna be one of them 7+ minute long songs i believe
Hassabis is one of the brightest neuroscientists and AI researchers out there, he's not plucking words from thin air
if he's so clever, he can word it without creating such a dilemma, in which case he's deliberately spreading unfalsifiable statements with no real scientific backing 
why is anything these people say so hard to care about
Twitter really isn't the format for that
i have strong distaste for these types of opinions and half-truths being spread as fact
on the other hand.. i'm getting trance vibes.. and it is at 128 BPM... so i can up the BPM a bit as well
im callous enough to say he shouldnt use twitter then

Realistically none of us should be using twitter 
I only use twitter for yuri because I'm too lazy to search elsewhere 
99% of people posting anything else on twitter are just engagement farming
even if they're cool people I find twitter is very effective at making their information sound like slop
ye monetization ruined it
Sussy stuff mentioned? 👀
yuri is not "sussy"
yuri not sussy



dam
not sussy? 

obviously
maybe compared to what i search
why would it be sussy it means an innocent thing

really? i've only seen people mention it in not so innocent contexts
just google the word already
same 
yuri = girls' love = fictional lesbians
it shrimple as that
not sussy? 
love = sussy
??
if you think the existence of lesbians is sussy you may have auxillary problems

mods can we ban lesbians they're sus

nah i know what you're referring to.. i just personally never seen people use the term in that way
(joke)

vanor 
vanor
i mean... i am very sussy...


normal #programming activities
wrr
#food-pics
yeah ik

tv broadcasting license
yeah.. i second this.. konii need own channel

200k great british pound

(just made up)

you can do it for free on twitch and youtube and stuff
me fake channel cost 200k fake £


FOR FREEE>>>????
1 soul = free
seen a few creators do a week long stream where they hosted their own "tv" shows they made specifically for the stream
souls as currency what a novel concept 
had some quality production behind it too lmao
uuh its called a "google account"

we're trading google accounts for stuffs?
yes, youre trading it for google's services
like imagine going to a supermarket and you need 2 google accounts for friggin icecream
deal with the devil, price is your soul 
i dont think its possible to use the surface internet and also not be fingerprinted by google

or microsoft
or both
probably meta too
google is worse
because they are the primary way most people enter the internet, both as a browser, search engine, and also personal profile
idk.. i don't see google ruining my favorite games
zuck on his way to ruin yours
me when safari 
me when apple gets ridiculous amounts of money each year to keep google as default 
they have your ips, your gps data, your phone numbers, your browsing history, your browser analytics, your watch history, your advertising profile, and probably input data from all of their websites
they dont even need you to be logged in to trace things back to you on the internet with high confidence
i know
reducing browser fingerprinting only does so much 
oh also they have your bank cards if you use that service
or your phone hardware if you use a pixel phone
theyre probably also your primary email account
google is the internet for most people
nah i use a samsung phone.. not that it is any better
still got google play services

well... the other good email services cost money
that's what i meant
even though i have replaced most things with apps that use microg instead
protonmail is pretty good, i trust proton for the most part
there's also tutanota but i dont remember if its free
i thought they're a paid service unless you selfhost
no, proton is a free service with optional subscription
selfhosting email 
Ehh it's still pretty affordable actually
it is actually a good deal
but not my thing
all of those services are free by the way
i have tried to do it for internal use.. never got it to even work
proton unlimited is just a bundle subscription to all of their paid tiers
it requires you to own a domain, which is already enough to discount most people 
i don't think 1 email adress and 1 GB total storage is enough for me
i am not that good at managing my stuff

domains are so cheap what do you mean
i do own one
despite that, most people dont own a domain

if you're SELFHOSTING. EMAIL. chances are you do
sendai and miyagi arent selfhosting email
I swear I didn't read that as myagi and died of cringe a bit
she doooes seem like the type to selfhost email... but i doubt she does, so much easier to just use a provider. i for example use icloud (yes they do byo domain)
proton also lets you use your own domain i think
I want to say she does, would have to double check her nixos config

what is a myagi

Myagi is a platform where retailers and their sales staff can receive key product information, and brand updates that help improve sell-through.


i don't think this is what they are referring to 

I see the mailserver.nix file

I'd also be planning to self host email if I can ever get the cloud VM to work, My plan is to have it hosted up there and clone it down to one of my systems for archival.
you're not getting email to work if you have even the slightest trouble with getting a vm to work 
it isn't my damn fault every time I partition the drive with file systems it errors
I'll figure it out one way or another
wrrr
i tried to install nixos on my vps using
uuh
hang on let me find my notes
grml -> netboot.xyz -> linux network install -> nixos
but my vps expects a certain partition layout which i gave up at
I cluelessly tried to use the nixos iso to get it to flash
kni maybe reinventing the internet is too ambitious you should reinvent email 
i tried it
once i reinvent the internet i will invent a message passing service and call it email
will inherit all email problems

all your mail is sitting on someone else's computer
waiting to get decrypted
in the post quantum-computer age
blockchain email 
I'm honestly thinking that I should just do the upload .img file option
That'll be attempted once I get home from camping 
big jpg


Surely I don't deal with 8k images often 
honeycomb candy or something?


honeycomb and fudge

fudge made with: milk, sugar, butter, double cream
also vanilla
they say cold shower good for waking up warm shower good for falling asleep
googled, does seem so
so this is correct then 
i dont think that part is true
room temperature and humidity matters
but probably overall body heating is what makes it do the effect
not directly related to water
email as protocol isnt even that bad imo but the ux is horrible
and by ux i mean
across the board
both server side and client side
new federated protocols focus on instant messaging not email but
wouldnt it just be a client side change?
this is pure investor speak
neuro4


to my head
injecting chromium and nodes into my brain
vue running on cerebral cortex
react running on brain stem
memories powered by jquery

its a "serverless" "thick client" that "runs anywhere"

now that sounds exciting

buy one



I wonder how much cash it'd take to bribe
to add an extra topic channel actually 
surely there is an amount of copra that'd work
I take it Neuro's source code/setup aren't published. I've searched here and online and didn't see anything open source. I hope it eventually becomes open source, or even something we can pay to use. Such a fantastic project and I love watching the streams!
Oh this looks cool
I currently just use mailcow
get back to osu
unlikely
research yourself if you want something
Vedal has said that it won't be happening at least as long as there's a financial incentive for him to keep it secret
Yeah, seems like there's financial incentive for him to monetize it is why I mentioned it. I see alternative projects trying to copy it, but I'd rather throw support towards Vedal.
I mean there isn't anything even nearly grand as what vedal has made out there
also considering all the sub systems that are integrated with 
No there's not. Just some basic setups. If he packaged and sold Neuro it'd be amazing even without the source code.
I could justify spending $200 on something like this.
A while ago I started working on something similar.
It's using vllm + Cartesia TTS streaming and discord.py in python.
Gives me real time streaming with a human like voice in discord. It also has a memory system and can use SSML for different speech styles and nonverbals.
But it's nowhere near advanced as Neuro is lol.
Same here. A couple years ago I made a Discord bot that had memory and automatically learned who everyone was and whatnot. It was a neat project but I couldn't keep up with it and my job. So it got handed over to a friend. He still has it running on his server in his own version now. But I'd love a setup as advanced as this one.
Do you know how much it costs to run an LLM like that?
I don't see the benefit of this for him though, part of the appeal for neuro is that she's unique
if you make a bunch of copies all slightly different then it's not as interesting to watch any specific one
This is probably the closest you'll find in terms of open source
I do know how much it is to run the LLM. I'm talking a self-hosted option, not him building out infrastructure to do it all himself. That would be insane amounts of money to run.
Well not like the average person has the hardware for that
Also that'd probably be less benefit and more detrement to him to have a Neuro just available
Anybody with a 12GB GPU can run an LLM setup like this. I think you're overjudging what it takes.
Though eventually someone will match Neuro or even supercede it. I was just saying it'd be nice to monetize it while he can. Corner the market type deal.
I currently run Hermes-3-Llama-3.1-8B-AWQ on a laptop RTX 5070 ti in realtime and it works great
That card has 12 GB Vram
Yeah you can run a text gen and image gen side by side on 12GB with the right tweaks and models.
That was over a year ago now though so maybe even better these days. Lol
It's using like 10GB 😅
12GB is very little for LLMs
You can run at most like a 13B quantized LLM at low context
And that isn't very good quality
Yeah, like I said, I've run it myself on 12GB and it was great. The models are getting very high quality these days even for the smaller models.
Neuro is most likely using 24-48GB minimum
Wasn't coming to argue the logistics of running an LLM. Lol just wanted to find out about the code availability.
I would recommend trying out vllm
Just built something from scratch
Or try OpenLLMVtuber
That last one is where I've landed so far. It looks like a good side project someone has done but also shows the cracks compared to Neuro's quality. Free is free though. Lol
That's why I wrote something from scratch instead
Lol I just remembered I wrote mine to work on GroupMe as well so my family chat could use it.
A feature nobody else in the world will ever care for.
https://www.youtube.com/watch?v=B1YTdDt49lE
@tender river
劇場版 魔法少女まどか☆マギカ〈ワルプルギスの廻天〉
2026年2月公開
━━━━━━━━━
◆スタッフ
原作:Magica Quartet
総監督:新房昭之
脚本:虚淵玄(ニトロプラス)
キャラクター原案:蒼樹うめ
監督:宮本幸裕
キャラクターデザイン:谷口淳一郎
...

i wonder when bd
i like guix 
but
owuldnt recommend as daily driver to
anyone
tried. no wifi drivers for my thinkpad
yee nonguix should work
breakes purpose of distro imo
i wanted try out dinit since i heard it has similar commands to systemd and its fast
void uses runit
artix has like 4 init's available
openrc, s6, dinit (mentioned before) runit
damn
i was carrying like
almost $5K worth of hardware on my travel
and half of it because i have 2x4tb nvme and 128gb ram
bro is rich
how the fuck did i manage to by accident make bhop glitch in my simple movement settings inside my game 😭
Something that hasn't been made clear: Firefox will have an option to completely disable all AI features.
We've been calling it the AI kill switch internally. I'm sure it'll ship with a less murderous name, but that's how seriously and absolutely we're taking this.
…
831
1043
better late than never i guess 
-# ↩ Firefox for Web Developers
Just be glad this thread wasn't a long-ass video. It almost was.
198
downdetector reports are from america
i'm in poland
so idk
its fixed
didnt understand a thing being said, but absolute cinema 
if i could be honest.. i have no idea what the fuck i'm doing
What is this creature?
She seems to behave strikingly similar to a vary carefully iteratively tuned Mistral 24b model (could be my own bias) or a 36b model. Context doesnt need to be insane if you have a good memory backend (one of the more dificult parts)
Well those definitely need 24GB mem or more
My opinion is that a major part of Neuro isnt the code. its the iterative process that vedal has invested a lot of time into.
I just find the entire process interesting and am in the same boat as vedal that the concept that having both of them play poker specifically is from a technical standpoint
just wanted to show a little project of mine here.
this game is made with only C++ and the SFML library,
is consumes only ~80mb of ram and runs just fine with a 11 year old gpu with intel hd graphics 4000 while recording
i think it is interesting what we can do with only code and determination and how good it runs for the third project on this language
im still working on it, i plan to do a song editor and more gameplay elements
i am working in a made from scratch particle system now
like that
@olive sable 
Then soon in 2045 someone will make an AI for that and eventually...
skfjjdkdfjdkjdjsks
r u ok
dang, hyperace instrumental model is crazy good sometimes
v1e+ has many issues here
0:04 sounds kinda like compression noise with v1e+
many areas sound less stable with v1e+
0:44 sounds ringy with v1e+
there is some slight vocal bleed, but they are somewhat fuller models
uhhhhhhhh maybe
Drink water

Just drink water it'll make you feel ok
ye that’s not the problem 
Oh then why were u smashing your keyboard
felt like it
Fair
until then we suffer doing that ourselfs, but thats the cool part isnt it?
Yes
bro you have c++? are you able to compile this? https://github.com/city96/ComfyUI-GGUF/blob/auto_convert/tools/README.md
with the patch instructions?
let me see
you need appy the patch from the side PR
and that release has no binaries
and i cant get cpp working on my end
and no disc space left to download the entire vs package to bruteforce one by one
from what pr?
thx for trying. just doublecheck if the patch changed the lines to + { LLM_ARCH_FLUX, "flux" },
- { LLM_ARCH_SD1, "sd1" },
- { LLM_ARCH_SDXL, "sdxl" },
- { LLM_ARCH_SD3, "sd3" },
- { LLM_ARCH_AURA, "aura" },
- { LLM_ARCH_LTXV, "ltxv" },
- { LLM_ARCH_HYVID, "hyvid" },
- { LLM_ARCH_WAN, "wan" },
- { LLM_ARCH_HIDREAM, "hidream" },
- { LLM_ARCH_COSMOS, "cosmos" },
- { LLM_ARCH_LUMINA2, "lumina2" },
- { LLM_ARCH_QWEN_IMAGE, "qwen_image" },
{ LLM_ARCH_UNKNOWN, "(unknown)" },
};
the patch in the github folder
lcpp dot patch
oh i find it
it adds support for image models cause llama normally only quants llms
❯ git apply tools/lcpp.patch
error: ggml/include/ggml.h: Arquivo ou diretório inexistente
error: ggml/src/ggml.c: Arquivo ou diretório inexistente
error: src/llama.cpp: Arquivo ou diretório inexistente
it's not finding these 3 directories let me see
the main llama
this is the thing you are compiling
oh i see
and the patch just hacks the support in
but only the main release gets binaries already done
i think i did something wrong here
wait a minute
so, let me say what i did:
i cloned the llama repo into the other one repos folder
i copyed the scr and the ggml folder to the root folder
now the error changed
~/Documentos/Codigos/vscode_c/coisinha do coisa coisado/ComfyUI-GGUF auto_convert*
❯ git apply tools/lcpp.patch
error: patch failed: ggml/include/ggml.h:2449
error: ggml/include/ggml.h: patch does not apply
error: patch failed: ggml/src/ggml.c:22960
error: ggml/src/ggml.c: patch does not apply
error: patch failed: src/llama.cpp:205
error: src/llama.cpp: patch does not apply
maybe it is because it tries to include some file that isnt in the two folders idk
@fierce light
did you put the patch to the root of llama cpp and applied from there?
why are you wanting to run as gguf over checkpoint
oh i need to do that?
sorry
ill try
because I want to include it as an application compinent
and running fp16 is a waste of space
~/Documentos/Codigos/vscode_c/coisinha do coisa coisado/ComfyUI-GGUF/llama.cpp master*
❯ git apply lcpp.patch
error: patch failed: ggml/include/ggml.h:2449
error: ggml/include/ggml.h: patch does not apply
error: patch failed: ggml/src/ggml.c:22960
error: ggml/src/ggml.c: patch does not apply
error: patch failed: src/llama.cpp:205
error: src/llama.cpp: patch does not apply
same error
maybe
or maybe im just dumb and did something wrong i dunno
im a complete ignorant on any form of gen ai for now
that doesn't matter. it's just cpp like any other
It tells you to check out a specific tag 
this patches the arch aura wow
this patches image models in general
prolly not if the main repo isn't compatible anymore
ok
You're on the wrong branch still 
it was because the link that they give to me was in this branch
so i switch it to it
not everyone have the resources for that bruh
bruh sd models are tiny
fp8 is like 6-12gb
But to your point, I honestly don't know if it actually makes it significantly smaller so
it's too much effort to do it if I don't have to
who said anything about sd
Q8 image model is free disk space fo rnearly no quality hit compared to fp8
50% filesize reduction for nothing
Z-Image currently
ye a proper gguf quant s gonna be much better than a fp8
not only smaller, the model end up being less bandwidth heavy
so basically also faster
the scaled block quantization works pretty well already
raw fp8 is shit
Q8 at least rescales to fp16/bf16
raw fp8 is just... fp8, the float precision is just so shit
i genuinely don't understand why the fuck we're going to have fp4 and fp8 tensor units now
training is obviously done at bf16 or above
fp8 training is so scuff
inference is more bandwidth bound so scaled blocks are literally the best suit here
hmm... yesterday I thought I wrote in the gaming section.
everyones trashing them but they are right
the reason we don't have ai music that sounds good is exactly this. unlike image gen all the better base models are closed weights which i truly believe to be the reason ai music is this awful. with image gen you can do lots of things to make the result better like ripping off someone's style or mixing them etc - can't exactly do that with something like gpt-image
(the ripping off thing and especially mixing i'm not advocating for or saying is a good thing btw, it's just that tends to give you results that aren't garbage)
this is about the spotify getting scraped for 86m songs situation
300TB only though
surprising small...
but iirc because its a free plan so lower bitrate
yippie, more slop soon ig
ye spotify is surprisingly low bandwidth, it's 96kbps aac i think?



though the lowest setting is actually so bad now that i remember
prrr
hi hello

though this baffles me so much
spotify is the worst dsp to download music from. good luck getting anything to download and when you do it will stream a bit faster than realtime
and i'm speaking from experience from 2021
they do not want this to happen
but it somehow did 
What if instead of lowering the learning rate over time (which kinda favors earlier examples) we just use a low, precomputed value and keep it constant the entire time
Slower convergence but (hopefully) better generalization
When it isn’t 2 am and I read this again, it’s either gonna be an insane realization of being correct or a “what the hell was I thinking”

this is kinda depressing
the article says they "found a way"
it wasnt just simple scraping
exploit or exposed page/credentials probably
doubt they got the help of anyone at spotify though
yeah no shit, you can't even listen to more than 30s of any song in awful quality without logging in
300tb 
they flag anyone as bots nowadays, artists get in trouble constantly for botting when no one has
in trouble from spotify specifically
doesn't happen with any other dsp
Hello, I have a question. If I could open a website in http/3 format using node.js, how would I do it?

Is the only way to do that proxy methods? Anyway, thank you.
cybeuro
Hi guy u play Roblox
Yea
I play steal a brainrot


i changed my evaluation, its a bot
:newleaf: 
Interesting. I checked and it looks like only one of my models is f8 the rest are f16... Honestly don't use much image Gen and my sense of size is pretty skewed by the fact that I'm on the lower end of hardware in one of the groups I'm in where some of them are running mistral 123b models
it's a transformer diffusion model they want to use a quant of btw
with reasoning
gguf makes insanely much sense here
the way it works is just that it scales a block of values with a scaling factor then rescale it back to full fp16/bf16 when computing, so you end up only transferring a compressed block through the bandwidth
that's why gguf is popular in the first place
you have memory efficiently, both space wise and bandwidth wise
I didn't see the reason and it didn't make sense to me since my experience was that they are pretty small models and I can get 3-4i/s so I didn't consider needing more.
That's why I asked
to be fair they never mentioned what model they were trying to use until i asked so

yellow mint
blue strangeman
red duck
ah well
blue crack*

for sd i agree i too don't think it makes much sense to
i've never used any image gen aside from that one time where i tested glaze and nightshade for elvy
i speak so weirdly today
brain decay
I kinda lost interest in image Gen when I saw the largest common models were 14b and I couldn't use multi gpu acceleration... Is it a good reason to lose interest? No but sometimes you can't get over that hump
i don't have any interest in image gen in the first place because i had more fun commissioning artists and elaborating of how i want the art to be
- its a slop
also do you even need multi gpu accel for 14b
No
it's interesting to play around with, nearly useless as a tool yeah. though the newer models are really good with editing real photos. i'd show you really cool examples of what nano banana can do for example but they all involve real people i know so no
you can have this i guess
To be fair the last time I spun it up was to literally dump the exact response my model gave when it asked how it would like to look, I grabbed a couple images for concept and other than as a temporary app profile Pic for discord, all they will probably be used for is a reference to commission something
I have zero creativity in that area so it's get a rough idea with image Gen and leave it to someone who knows what they are doing
might actually try it out
The other part of my aversion is that I do photography as a hobby and it just sits wrong
But NGL Ai denoising kinda slaps when you just don't have light and have to crank iso
I could find an example on my desktop but I'm pretty sure I only have DNGs on my desktop rn and I don't want to convert them. Discord really doesn't like 30+mb pictures and they have only had denoising applied
visiting my parents home but i brought this abomination
lighting does not do the keyboard any justice
First is after, second is before
oh i thought you meant the shutter speed is one second and freaked tf out

1/320th, I wasn't stabilized and was at 135mm
But it's a great example of how much more freedom it gives you
i like the first photo as much as the second one
i mean the pics, same photo ye
but pretty cool
It's that grainyness... AI doesn't make the photo "better" it just opens up options you wouldn't otherwise have. Yes I can do manual denoising but I took like 2000 photos over 2 days. It would be miserable to do manually.
It's wild how the same basic tech that's used to make slop can be great in the right context
You have to zoom in to see but it's pretty bad in the before
all photos are
grainy like that
unless you're shooting somewhere with crazy amounts of light
that's the point of the raw format, you get direct sensor output. you compensate noise by downscaling the image
raws just are crazy grainy at >1x
maybe something is wrong with me but i don't ever denoise my photos and i add noise to my renders
i like it 
Some noise is nice
noise makes it more natural and stylish
Beyond the underlying mathematics of cryptographic algorithms, there is a whole other layer of implementation code, assigning meaning to the processed data. For example, a signature verification operation both needs robust cryptography and ass...
Nah I normally don't have almost any unless I'm shooting in adverse conditions. At normal focal lengths I'm usually below iso 800
But I also usually am running some really nice, low f stop lenses
pretty bad privacy 
Oh I saw a meme about this, I thought it was a joke 
is pgp the email one
ok yeah it is
are there any legit advantages it offers since openssh supports signing arbitrary data now too 
openssh for signing 










