Andrej Karpathy comes out as a C supremacist, implements an LLM training stack in pure C while explicitly rejecting Rust and declaring C the most beautiful language
Rustsisters...our response???
Andrej Karpathy comes out as a C supremacist, implements an LLM training stack in pure C while explicitly rejecting Rust and declaring C the most beau...
lisp wizards unite to resurrect our true savior the Lisp Machine and prove all these heathens wrong, make them confess their sins and kneel before the One True Language
Holy-C superceded lisp
Absolutely based.
>builds a custom processor to run lips
>slower than a generic risc microprocessor running lisp
What did they mean by this?
we need a lisp without GC, then it's over
interesting
https://github.com/karpathy/llm.c
he said later in the twitter thread that he's in the middle of porting it to CUDA as well
Jesus Christ that is some beautiful code
>residual_polygamic_backward_layman_enthropy_calculator_run_gnu_linux_impl_v_212171172732632632( B*S*T/X+Y/Z+(A-B-C) );
DISGUSTING
thats basic stuff there are quite a number of those projects on github some even implemented their own GEMM library instead of this basic stuff here. Its not hard at all. If anyone wants a good resource read up on https://en.algorithmica.org/hpc/ . There used to be a very in depth blog series by one of the blis devs but I can't really find it its worth checking out since the hpc stuff isn't that advanced.
Anyway I think a ISPC or taichi implementation would have been far more elegant there is no need to reinvent the wheel again he has done it before already.
Holy shit this code is verbose. Why then do Cniles say that Rust is bad because verbosity, if their language is verbose as well?
because you touch yourself at night
the problem with rust is bloat, not actually writing code in it
literally nobody cares about verbosity you psuedo-coder
What would you remove from rust without sacrificing some of it's performance or safety?
Go is a managed language lol. Why are you bringing it up?
>safety
>performance
a moronic truckload of libraries that do literally nothing and if anything are probably malware, id remove that
go ahead and see how big hello world is in rust, if the language/platform was so smart why does it package incompressible shit that literally isnt needed?
>go ahead and see how big hello world is in rust, if the language/platform was so smart why does it package incompressible shit that literally isnt needed?
You have no idea what you are taking about.
thats big as hell for hello world are you kidding me
kids these days should be forced to learn on c64s I fricking swear
It's literally less than C, compiled with gcc -O3
Try using gcc -s -O3, might (not) be a fair comparison, idk.
Any direct performance/speed comparisons with the Python implementation?
noobs. pic rel
OH NO NO NO NO NO
KEK
>.unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap().unwrap()
more beautiful than troony code
Delusional
>look at C code
>readable
>look at rust code
>readable if you know rust
>Shows cplesples gibberish
oh my you sure pwned me gud!
Being explicit isn't a bad thing, it's the other problems that are specific to Rust.
Like the readability, if you haven't put in 400 hours you're going to struggle to understand everything being specified.
Golang can be pretty verbose but it's easy to read.
Another complaint is you can't really choose to skip the explicit verbosity to go fast in Rust which is good in you can't be lazy but bad in that you need to already have solved the problem in your mind fully before you tackle it in Rust.
It's hard to test, discard, and implement new ideas in Rust when you're having to specify everything each time.
Is it verbose with bullshit "intent" or verbose with things that are actually executed?
Actually executed? What ISA has a #define instruction again?
no one's ever claimed that rust was verbose you closet troon, in fact rust syntax is as shit as it is precisely because it tries to be so terse.
>no one's ever claimed that rust was verbose
NTA but this is absolutely a claim that has been made many times right here on this board
no it hasn't. post examples or have a nice day troon. I'm not a mind reader
>there is no need for python
>start
>`python whatever`
Why are cniles like this?
He states it's impractical to train your own data so you need to use the py shit for now.
>There is no need for 245MB of PyTorch or 107MB of cPython
Based.
>direct communication with the machine
communion. i know, it is a hard word for the average rust user.
you don't know what that word means, ESL
Yes he does, simplifying it to just "communication" loses the "Praise be the Holy C, Machine Spirit and all it's Binaries" vibe
No he doesn't.
Black person, do I have to quote dictionary at you? Are you mutts so moronic that you even forgot how to English?
I'm not a porkmutt, you're the pork!
We speak AMERICAN here, thirdie
it's a latin word you stupid moron. literally everyone who speaks a romance language knows what that word means better than a dumb mutt who only knows it as a loan word.
Modern English is both a Romance language and a Germanic language. Anybody who knows English as well as one language from both of these families knows exactly what I mean, and it has nothing to do with vocabulary.
What's German about it?
German as in High German as in the language of Germany? Nothing. But that's different from *Germanic* anon.
English's Germanic roots shine through in the fusional quality of its nouns, as well as the types of affixes and inflections applied over them.
Refrain from speaking if you don't know what you're talking about, anon.
>dumb chink never heard of the word communion
>thought it was “communication”
Kys chang
C Is Not a Low-level Language
Your computer is not a fast PDP-11.
C is definitely a low-level language, or at least can be, lol. There are people in embedded who only touch C over the course of a decade or so.
There is no hard definition. People used to call anything that's not 1 to 1 to machine language higher level.
https://en.wikipedia.org/wiki/Low-level_programming_language
C Is a Low-Level Language
Your computer is a fast PDP-11.
C is low level language when you compare it to Java, python and Javascript. Assembly is lower, so C is considered high level language when compared to it. Then there comes binary, which is even lower than assembly. It's a matter of perspective.
It's not.
>A low-level programming language is a programming language that provides little or no abstraction from a computer's instruction set architecture
The word you are looking for is systems programming language.
>when you compare it to...
It is.
https://en.wikipedia.org/wiki/High-level_programming_language
>The amount of abstraction provided defines how "high-level" a programming language is.
C is not a "low-level language." But it is a lower level language than Java, Python, and JavaScript.
python is low level language confirmed
If C is low level, you can use Cython I guess. Does that make python low level enough? Honestly hate the words high/low level. It's basically meaningless. I'd argue any assembler are the only low level languages that exist anyhow. Anything else sugars over basically everything low level about the actual execution of the code, even if in practice, it's usually pretty predictable. Of course relying on this will result in a language lawyer troony telling you using it is undefined and your code won't do anything useful without some compiler argument, but whatever.
(me)
Low-high level language is kind of a moron system. Python is too far away from the computer's understanding language so it needs a lot of translating in-between for the code to be comprehensive enough for the computer to run thus resulting in slower run-time under normal circumstance. The closer it is to binary form, the simpler the compiler (the traslation machine) and the less the abstraction and the faster it is for the computer to run.
yes idiot i know what C is topkek
no it doesn't, C was the language of choice for decades, it will remain the choice for a very long time
It is more a fast PDP-11 than a Rust Machine.
debunked
https://www.yodaiken.com/2021/05/21/your-computer-is-a-fast-pdp-11-and-more-on-c-the-c-standard-and-computer-architecture/
>yodaiken
He is a UB denier, unironically even worse than anti vaccination morons.
https://www.yodaiken.com/2021/05/19/undefined-behavior-in-c-is-a-reading-error/
He's right though.
So now I know the source of the moronation expressed by rust trannies in yesterday's thread.
>https://queue.acm.org/detail.cfm?id=3212479
>Meltdown and Spectre vulnerabilities
OK
>Both of these vulnerabilities involved processors speculatively executing instructions past some kind of access check and allowing the attacker to observe the results via a side channel.
Good so far.
>The features that led to these vulnerabilities, along with several others, were added to let C programmers continue to believe they were programming in a low-level language
OH YOU FRICKING IDIOT
Speculative execution has NOTHING TO DO WITH C, nor the PDP-11, nor any particular language. It's about making effective use of superscalar pipelined CPUs. Take a 10 million cycle loop. You want a bunch of nops in your ALU at every iteration? Or do you want the CPU to speculate and get work done since it's going to be right 9,999,999x? Let me put it another way: do you want 2010s/2020s performance? Or performance like it's 1990?
>processor architects were trying to build...processors that expose the same abstract machine as a PDP-11.
As I tried explaining to the Rust trannies yesterday: the PDP-11 was not unique. Most of the architectural features of today's CPUs were worked out years before C or the PDP-11. They are the consequence of EE, not of C.
>This is essential because it allows C programmers to continue in the belief that their language is close to the underlying hardware.
No. All languages do the same things because that's how the machine works. Most just hide what they're doing from you.
>muh GPU!
Graphics are embarrassingly parallel and do not require all the instructions a CPU requires. Most problems are not and do. That's why threading real world problems is hard. Thread cost is due to context switching a machine with a relatively high level of state. Again: solve any problem vs pixels go fast.
>So now I know the source of the moronation expressed by rust trannies in yesterday's thread.
anyone who mutters "muh pdp 11" is almost certainly parroting that idiotic article brainlessly. C does have many problems, but nearly nothing that article huffs about is in anyway related to C.
the reason dunking on C is popular among academics ever since it's inception is the same reason UNIX is disliked by them - it was so widely successful that it basically killed any academic OS research (which were usually highly complicated due to academic masterbation) https://dreamsongs.com/RiseOfWorseIsBetter.html
C & UNIX causes academics to seethe hard into chimpmunk rage because to them, it's a reality check. a cruel reminder that 90+% of their over-engineered papers they like to circlejerk around is completely and entirely useless in the face of a simple and practical implementation. https://youtube.com/watch?v=k0qmkQGqpM8
>a cruel reminder that 90+% of their over-engineered papers they like to circlejerk around is completely and entirely useless in the face of a simple and practical implementation.
C shill translations:
"over-engineered papers" = one man implementation that could run on 1970s hardware
"simple and practical" = 10,000 programmers from 1000 companies
>"over-engineered papers" = one man implementation that could run on 1970s hardware
Multics, the academic OS project, was so overly complicated that the project never even finished. it had to be abandoned by bell labs.
>"simple and practical" = 10,000 programmers from 1000 companies
UNIX & C were programmed by 2 (two) people with ZERO support from bell labs because they didn't want to fund another OS project after the colossal failure that was Multics.
>"over-engineered papers" = one man implementation that could run on 1970s hardware
Multics, the academic OS project, was so overly complicated that the project never even finished. it had to be abandoned by bell labs.
>"simple and practical" = 10,000 programmers from 1000 companies
UNIX & C were programmed by 2 (two) people with ZERO support from bell labs because they didn't want to fund another OS project after the colossal failure that was Multics. https://en.wikipedia.org/wiki/History_of_Unix
despite your post containing absolutely no political statement, i can still tell that you're a libtard. how? because stating something obviously and observably the OPPOSITE of reality with a straight face is the prime hallmark of a libtard.
>MULTICS was the only alternative to UNIX
WTF even is this post. Nobody brought up MULTICS
And no one is giving a concrete example of a better alternative and why.
Everything else in 1969 was superior to UNIX in every way except for the fact they were expensive, proprietary (so was UNIX but everybody ignored this and shared source anyway), and ran on expensive machines with more memory than a PDP -7. UNIX is a virus that got way out of hand, especially when everyone's mother and their dog was making their own processor and needed a cheap OS to run on it. UNIX was basically the only option unless you wanted to pay $$$ for something reasonably usable. The only other plus side to UNIX was it booted fast, which was a major plus given the fact it crashed constantly.
>Everything else in 1969 was superior to UNIX in every way except for the fact they were expensive, proprietary (so was UNIX but everybody ignored this and shared source anyway), and ran on expensive machines with more memory than a PDP -7.
Concrete means specifics, not a blanket naked assertion. And as I pointed out in another thread, the assertion that 'UNIX was chosen because cheap machines needed a lightweight OS' is false. You say this as if UNIX was running on a fricking VIC-20. Early UNIX workstations based on the 68K line had performance in MIPS as well as RAM comparable to early 1970s minis/mainframes. And the 68K ISA had the required features. They were powerful enough to have run the competing OSes. By the time RISC workstations appeared they definitely were powerful enough to run any contemporary OS. Hell, hosting a GUI was more CPU intensive than the guts of the OSes in question which is why Macintosh needed the 68000 despite having an otherwise primitive OS kernel hand coded in assembly.
>UNIX was basically the only option unless you wanted to pay $$$ for something reasonably usable.
Early workstation vendors typically did pay for it.
>as I pointed out in another thread
LOL! You really do live in your own head
Two very similar conversations are going on, it's not unreasonable to think that people are reading/replying in both.
Pretty much spot on.
Nice follow up examples.
CPUs are designed to run the old cruft a little bit faster than the prior iteration. They're cruft processors, not general processors.
>anything that's not my dream academic machine is "cruft"
See:
The guy who wrote that article is a moronic homosexual.
That is if you are talking about instruction caching and pipelining. If you care about that stuff then of course you are not talking directly to the CPU when writing C/assembly since you have to take more things into consideration. But there is no language that does that, so it does not make C any less low level in comparison to languages like Rust.
Implicitly calling out that rust will NEVER be a real programming language.
>C so nice, simple, clean, portable and beautiful, aesthetically
>direct communion with the machine
April Fools!
>cuda
why not opencl?
but cool regardless
>c is close to the machine
Why do people keep saying this?
>asm is close to the machine
>actually the CPU just goes shit on its own
Why do people keep saying this?
>my Intel CPU computer is running Windows/GNU Linux
Why do people say this? Your machine is running Minix, all Intel chips are.
AT&T Hobbit had a massive influence on CPU implementations.
>AT&T Hobbit had a massive influence on CPU implementations.
Really? Name a commercial, shipping CPU that implements one or more of its unique features. Not its RISC features, but the features it has which are distinct from prior commercial chips, CISC or RISC. The features that make it a "C processor." Because I can't think of one.
They're moronic children trying to virtue signal their way into looking competent
The code would be 10x more readable if it were modern C++.
>modern C++
let's not kid ourselves
also, how do you define modern? there are like 5 different revisions of C++ from the last 15 years
Modern as in all the features implemented in the gcc/clang master branches
I define modern as using C++17 or newer features when possible, or at the very least, C++11.
>nice, simple, clean, portable and beautiful
Integer promotion / conversion rules already show that all these are wrong.
What did andrej invent? Did he invent VJEPA which predicts the missing portion of a video?
I will write the same thing in a single heavily templated .hpp to btfo him.
i'm more of a .hh man myself when it comes to constexpr abuse
what's this new homosexualry of chromatic aberration on twitter posts?
Who?
Candle is a minimalist ML framework for Rust with a focus on performance (including GPU support) and ease of use
https://github.com/huggingface/candle
Imtrans btw
Watch him get banned and ostracized for "transphobia". He's a dead man walking, simply by posting that
ah a thousand lines of pure joy
>LLM garbage
I can't begin to describe how much I don't care.
>somebody writes something in C
>rust trannies garther to seethe about it
More like
>rust exists
>ciniles gather to seethe about it
We would never think about Rust if you didn't bring it up.
Fricking this.
>see thread
>someone mentions c
>REEEE WHY CAN'T YOU RUST YOU FRICKING CNILE???
Have you even seen the OP?
No what does it say
As expected
That refers to all the Rust threads, nigtard.
Majority of Rust threads are made by seething ciniles. What's your point?
No they're not. Patently you are lying.
qed I win
Every programmer that moves from Rust to C increases the average IQ of both groups
That doesn't make any sense
it implies that every rust user has higher iq than every c user but only the dumbest rust users would transition to c. it's not that hard to make sense of, even if it is false.
Sorry I'm low IQ
you posted a crosspost and 5 deleted posts. good job actually doing the bare minimum after being explicitly told to like a child but you have no right to be giving me attitude. slit your throat
those anons mean verbose as in bloat. the syntax looks messy as hell because rust tries to do shit they don't need it to. they don't want the pointless tedium of jerking off to the algorithm they invented just to appease the borrow checker. but only a moron would see this and think the takeaway is rust isn't code golfing trash with its typing system and various other constructs.
>t. what is the archive?
And since when are the Cniles troons?
It's Cniles vs. Rustroons, stay in your lane.
Since 1987. See Mary Ann Horton. All of the transgenderism at tech companies started at Bell Labs.
>I didn't LITERALLY mean the exact words I said, and NOBODY would take them at face value
Completely valid and reasonable behavior and a great way to make an argument
I won't even respond with a (You)
>{asthetically}
This man is clearly on crack cocaine
If he wants to truly assimilate with the machine god why isn't he writing assembly?
Just use D, D is C with opt-in memory safety, GC, and literally the world's best templating/metaprogramming facilities