https://justine.lol Justine Tunney's Web Page https://justine.lol/lex/ https://justine.lol/lex/ Thu, 31 Oct 2024 23:59:59 PDT I just learned 42 programming languages this month to build a new syntax highlighter for llamafile. I feel like I'm up to my eyeballs in programming languages right now. Now that it's halloween, I thought I'd share some of the spookiest most surprising syntax I've seen. https://justine.lol/mutex/ https://justine.lol/mutex/ Wed, 02 Oct 2024 08:28:00 PDT Imagine you have a workload where all your threads need to do a serialized operation. With Cosmo, if you're looking at htop, then it's going to appear like only one core is active, whereas glibc and musl libc will fill up your entire CPU meter. That's bad news if you're running a lot of jobs on the same server. If just one of your servers has a mutex bloodbath, then all your resources are gone, unless you're using cosmo. It's still a new C library and it's a little rough around the edges. But it's getting so good, so fast, that I'm starting to view not using it in production as an abandonment of professional responsibility. The C library is so deeply embedded in the software supply chain, and so depended upon, that you really don't want it to be a planet killer. If essential unquestioned tools are this wasteful then it's no wonder Amazon Cloud makes such a fortune. https://github.com/jart/cosmopolitan/releases/tag/3.9.2 https://github.com/jart/cosmopolitan/releases/tag/3.9.2 Sun, 22 Sep 2024 03:41:00 PDT Cosmopolitan's Windows support may finally be feature complete. It's now possible to send signals between processes using kill() on Windows. Ten new torture test programs have been written to tease out more fixes and offer a high level of assurance that signal handling is correct. Some of these tests are good enough to deadlock the signal handling of UNIX OSes but not our signaling module for Windows. They also demonstrate that our Windows signal handling actually outperforms many UNIX OSes at latency. https://justine.lol/history/ https://justine.lol/history/ Fri, 23 Aug 2024 05:00:00 PDT In a world of infinite automation and infinite surveillance, survival is going to depend on being the least boring person. Over my career I've written and attached my name to thousands of public source code files. I know they are being scraped from the web and used to train AIs. But if I ask something like Claude, "what sort of code has Justine Tunney wrote?" it hasn't got the faintest idea. Instead it thinks I'm a political activist, since it feels no guilt remembering that I attended a protest on Wall Street 13 years ago. But all of the positive things I've contributed to society? Gifts I took risks and made great personal sacrifices to give? It's erased from the history books. https://justine.lol/matmul/ https://justine.lol/matmul/ Sun, 31 Mar 2024 17:59:15 PDT I just wrote 84 new matrix multiplication kernels for llamafile which enable it to read prompts / images faster. Compared to llama.cpp, prompt eval time with llamafile should go anywhere between 30% and 500% faster when using F16 and Q8_0 weights on CPU. The improvements are most dramatic for ARMv8.2+ (e.g. RPI 5), Intel (e.g. Alderlake), and AVX512 (e.g. Zen 4) computers. My kernels go 2x faster than MKL for matrices that fit in L2 cache, which makes them a work in progress, since the speedup works best for prompts having fewer than 1,000 tokens. https://justine.lol/oneliners/ https://justine.lol/oneliners/ Mon, 4 Dec 2023 09:00:00 PST I spent the last month working with Mozilla to launch an open source project called llamafile which is the new best way to run an LLM on your own computer. So far things have been going pretty smoothly. The project earned 5.6k stars on GitHub, 1073 upvotes on Hacker News, and received press coverage from Hackaday. Yesterday I cut a 0.3 release so let's see what it can do. https://justine.lol/cosmo3/ https://justine.lol/cosmo3/ Tue, 31 Oct 2023 09:00:00 PST After nearly one year of development, I'm pleased to announce our version 3.0 release of the Cosmopolitan library. https://justine.lol/sorting/ https://justine.lol/sorting/ Mon, 12 Jun 2023 09:00:00 PST A few days ago, DeepMind published a blog post talking about a paper they wrote, where they discovered tinier kernels for sorting algorithms.