jart

49 ms·

811.
by jart 3y ago
mmap() requires that you map at page aligned intervals which must be congruent with the file offset. You can't just round down because some gpus like metal require that the data pointers themselves be page aligned too.
812.
by jart 3y ago
Don't believe Fox News when they say this guy is a Google exec. It looks like he's just a product manager. I wouldn't hold being spicy on Twitter four years ago against any man, but someone torpedoed their Gemini product and
813.
by jart 3y ago
Oh that's probably why then. I'm allergic to anything popular. I'm blind to it like ad blindness. When I see it I'm like the way my cat slinks under the couch when the mail lady arrives. I like to use Twitter to hang out
814.
by jart 3y ago
What makes you think that? Which LLMs?
815.
by jart 3y ago
What is all this noise about bots? I have 36k followers on Twitter. I'm on it all the time. I don't think I've ever seen a bot. What exactly do these bots do and how would you know they're bots?
816.
by jart 3y ago
At Google they draw the distinction based on i/o. Anything that does i/o wouldn't be considered a unit test by them. Unit tests are usually for algorithms, math functions, data structure classes, and stuff like that. Most dev
817.
by jart 3y ago
LLMs will give normal people a firmer standing in technological society. That's a good thing. But will it change everything? Not a chance. Even if LLMs did change everything, that probably would not be a good thing. Dijkstra says Musli
818.
by jart 3y ago
I just declare firewall jubilee every now and then, where I flush the iptables and let people try again. It's also because people usually only control the IPs they use temporarily, so I don't want someone innocent later on to be b
819.
by jart 3y ago
What they did to me was flood the Linux Kernel with TCP connections. That's why it's so important to block IPs in the raw PREROUTING table. You need to nip it in the bud before Linux starts allocating any memory to the attacker.
820.
by jart 3y ago
I rent a GCE VM and there's not many if any people out there who can exhaust Google's network infrastructure. The only thing I have to worry about is making sure my server doesn't respond to abusive traffic.
821.
by jart 3y ago
I don't think Google sets the agenda. They get their orders like the rest of us, and then probably handed it over to an army of low-paid offshore contractors to implement. How else could you explain Gemma believing Abe Lincoln was blac
822.
by jart 3y ago
Use a token bucket on your web server to catch abusive IPs and then blackhole them using `iptables -t raw -I PREROUTING -s ip -j DROP`. I know. I run https://ipv4.games/ which invites hackers to unleash their botnets, and
823.
by jart 3y ago
This guy had a degree in computer science. He was an active member of the military who had served for three years. He knew how to code in Python and Perl. He had a bright future ahead of him. Yet he was so unhappy about the situation betwee
824.
by jart 3y ago
You'd first need to help them live with dignity. For the genuinely homeless the only social environment that's equipped to care for them are mental hospitals. Secondly you'd have to recalibrate your expectations for what prod
825.
by jart 3y ago
I agree. Also Mixtral is a heck of a lot more useful than GPT-2, which is the last thing OpenAI gave us before they went the other way.
826.
by jart 3y ago
It sounds like they took the Federal Reserve's business model and applied it to computing.
827.
by jart 3y ago
The government cleans the streets regularly, so relocating unclean people into places the city doesn't clean isn't going to make the city any cleaner. You probably haven't had much experience being around people who can'
828.
by jart 3y ago
> Legitimate criticisms include not working correctly out of the box for llama.cpp I don't think that's a legitimate criticism. Especially not for something that just launched. You should be helping your own project first, befo
829.
by jart 3y ago
Cosmopolitan is a fan of you :-) great work on gemma.cpp. I'm really impressed with it so far.
830.
by jart 3y ago
I'd wager the vast majority of Linux users have never encountered a memory safety issue with the Kernel. Memory safety issues are usually rare enough in world class C code that, by the time it reaches end-users, you have to actively go
831.
by jart 3y ago
Then they should say "we're researching how to create a memory safe language" or "we've built a memory-safer language" rather than saying what they're saying.
832.
by jart 3y ago
That list of core type system soundness issues is long enough that it'd probably be easier to prove Rust style checks are unsound due to incompleteness / halting problem issues rather than fixing all those issues.
833.
by jart 3y ago
But languages like C / C++ don't claim to be safe.
834.
by jart 3y ago
> unsound issues Apparently there's 84 open issues which the Rust developers consider unsound issues. https://github.com/rust-lang/rust/issues?q=is%3Aissue+is%3Ao... That's it. Rust's claims are
835.
by jart 3y ago
What is mempool? Is that this? https://en.wikipedia.org/wiki/Region-based_memory_management
836.
by jart 3y ago
It is.
837.
by jart 3y ago
True, but with some stuff you just ain't gonna need it. For example, chibicc forks a process for each input file. They're all ephemeral. So the fork/_exit model does work well for chibicc. You could compile a thousand files a
838.
by jart 3y ago
> The whole point of a good mempool is that you malloc once, and only call free when you exit the program So you're describing fork() and _exit(). That's my favorite memory manager. For example, chibicc never calls free() and i
839.
by jart 3y ago
malloc() already keeps track of every memory allocation. Just what kind of tracking are we talking about here?
840.
by jart 3y ago
Changing the way transformer models works is orthogonal to gaining good performance on Mistral. Groq did great work reducing the latency considerably of generating tokens during inference. But I wouldn't be surprised if they etched the

More ›