xena

64 ms·

511.
by xena 2y ago
I'd use Quassel, but I have custom weechat scripts. I can't easily implement those in Quassel.
512.
by xena 2y ago
I already have a Kubernetes cluster with distributed storage set up. Adding Proxmox makes that setup unique. I don't want unique. I want to reuse the infrastructure I already have.
513.
by xena 2y ago
Because I already have a kubernetes cluster with distributed storage. This way makes it somewhat standard with the rest of my setup. In SRE things the real enemy is when parts are unique in ways that conflict with the normal way of doing th
514.
by xena 2y ago
As a long-time Docker-slonker I agree, however then that requires me to remember to upgrade the container images which means that it's harder to ensure that it's up to date. If you set up a cronjob to update the VMs packages and t
515.
by xena 2y ago
Author of the article here. One of the main reasons is that I want to update weechat without rebuilding the container, taking advantage of weechat's soft upgrade method: https://weechat.org/files/doc/weechat&#
516.
My IRC client runs on Kubernetes(xeiaso.net)
137 points by xena 2y ago|127 comments
517.
by xena 2y ago
Hi, I'm the person that wrote that sizing comment in the draft for this article. I have been trying for a while and have been unsuccessful at getting 405B running on any of the GPU machines. I suspect I'd need a raw 8xA100 node to
518.
by xena 2y ago
Simply make rent, housing, and food free. Then people need not make money for the majority of needs.
519.
by xena 2y ago
I mean you can race the beam yeah, probably easier because you have two cores, but a framebuffer is easier for developers to work with.
520.
by xena 2y ago
The pro gamer move is to release it and then whoever copystrikes you is the owner
521.
by xena 2y ago
I'd need to do the math on this, thankfully most of it is on a GPU kind that I don't normally use. I probably over-scaled so that I didn't need to think about the scaling issues, but I can tally it all up.
522.
by xena 2y ago
To be fair, the Luddites weren't anti technology. They were standing against factories pricing them out of labor.
523.
by xena 2y ago
That leader was written in about 30 seconds because I thought that the space needed filling. I'm glad it works!
524.
by xena 2y ago
That's like changing the definition of UTF-8. You have no idea what is depending on the exact byte order of everything spooling out from the core.
525.
by xena 2y ago
I tried to but I had to remove it for rate limit reasons :(
526.
by xena 2y ago
I think I've scaled around the worst of this. Gonna go back to playing Zenless Zone Zero. Mention me on X if it breaks again.
527.
by xena 2y ago
Sorry! Adding more compute ASAP. Didn't expect it to get this popular.
528.
by xena 2y ago
Here's the prompt: Give uplifting words of encouragement for the following github profile. Use two newlines between paragraphs. Here are the details: Username: ${username} Name: ${name} Bio: ${bio} Company: ${comp
529.
by xena 2y ago
Bingo. Never thought it would get this popular and I am not using an architecture that autoscales elegantly. I'm adding more RAM and CPU cores to see if that helps.
530.
Praise My GitHub Profile(praise-me.fly.dev)
215 points by xena 2y ago|73 comments
531.
by xena 2y ago
(I work for fly.io, but not on Machines) Speaking as someone that has built experiments on top of zfs zvols ( https://xeiaso.net/blog/waifud-progress-report-2/ ), I'm not entirely sure what the main usecase of
532.
by xena 2y ago
I wanted to build something on top of Lemon Squeezy because stripe banned patreon-like services. Are you gonna extend that to Lemon Squeezy?
533.
by xena 2y ago
Where do you get access to those as a member of the general public?
534.
by xena 2y ago
Iosevka and Podkova!
535.
by xena 2y ago
Not if you want to keep the magic green compliance checkbox!
536.
by xena 2y ago
Every model has its own unique vibe to it. It's why new models feel better than they are.
537.
by xena 2y ago
The reason that companies align models is so that they don't get on the front page of the new york times with a headline like "Techaro's AI model used by terrorists to build a pipe bomb that destroyed the New York Stock Excha
538.
by xena 2y ago
Easy head math: parameter count times parameter size plus 20-40% for inference slop space. Anywhere from 8-40GB of vram required depending on quantization levels being used.
539.
by xena 2y ago
I moved to Canada in 2019. It ain't perfect, but it's home.
540.
by xena 2y ago
Yeah, the best videos follow the pattern of "What is this thing?" "Why does it need to exist?", How does it fit into the bigger picture?", and then finally "How do you actually use it?" This kind of stuff

More ›