GPU: NVIDIA GeForce RTX 4060 Ti 16gb (I typo'd the GPU above)
(This is via Ollama on Ubuntu.)
But 1-3 tokens per second is much faster than a lot of other high end models I've tried, so I was pretty pleased with it. Obviously other models run much faster on this hardware though.
qwen3-coder-next runs fine on my consumer grade nvidia 4070. Performance is not spectacular, but it's only a little bit slower than a properly-fit model.
Can someone explain what the current state of model benchmarking is? If you try to look up what the best locally runnable model is, you get a bunch of random blog posts using idiosyncratic criteria to rank things seemingly based on one dude's opinion.
Ideally I would love to see a leaderboard with relatively objective ranking criteria that 1. lets you filter by open weight / locally runnable, 2. filter by date of release (nothing older than x), and 3. is agnostic to hardware requirements. I just want to know what the best model is. Let me worry about how I will afford to run it.
I love the llmfit project for seeing what will run on your hardware, but it would be nice to know what I'm missing out on by not having better hardware, thus why objective hardware-agnostic ratings would be helpful.
For All Mankind aired an episode today that movingly commemorated the fictional lead character Ed Baldwin's Apollo 10-like in-universe mission on the same day that the real world Artemis II mission which also strongly resembles Apollo 10 landed safely. A strange and moving coincidence.
You're right, it absolutely does make sense. And yet it annoys me anyway.
There's been research by linguists (John McWhorter comes to mind) analyzing this phenomenon and it basically just comes down to the fact that as we age, we get more set in our ways, so the linguistic innovations that younger people do just have a tendency to annoy us, even when they logically follow or are objectively useful.
I try not to let it bother me, because it's irrational to feel that way, but it just does lol
Your reply did exactly what I complained about: expanding the definition of emulator to cover reimplementing a network protocol.
You're not wrong that "server emulator" is a generically correct use of the term emulation, in the same sense that it is a correct use of the word for someone to say they emulate a fashion sense of a celebrity they like in their own wardrobe.
But in computer science, strictly speaking, the original definition of emulator was more strict. It was about things like emulating processor architecture A so as to execute programs written for it on processor architecture B.
And part of why expanding the definition to include "server emulators" annoys me is why has this definition expansion occurred only in gaming contexts? If a free UO server is a "server emulator" then why is Samba not also a server emulator? The lack of consistency is irritating to me, and it only happened because gamers like the term emulator, not due to any kind of rigorous computer sciencey reason.
I'm a fan of UO and I love seeing more projects like this. Nice work!
Obligatory nitpicky aside, a time-honored tradition of HN:
I've long been irritated by the use of the term "server emulator" in gaming contexts. Technically these projects are just reimplementations of a proprietary networking protocol. Nobody calls Samba a "server emulator" because it reimplements the Windows file sharing protocol, because Samba isn't "emulating" anything from the perspective of the traditional definition of "emulator" in computer science.
But for some reason, I guess because "emulator" has colloquially been redefined by non-CS nerd gamer normies as a term for software that lets you play proprietary games on platforms they were not designed for, we have ended up in this new status quo where the term's definition has expanded in this game of telephone way that annoys mainly me and not many other people.
And what's kinda funny is I say that it is a "new" status quo, but it's not even that new. I recall, what, like 20 years ago now I was in an edit war on Wikipedia fighting with people over the "server emulator" article, insisting that the term was technically inaccurate and should not be used. Unsurprisingly in retrospect, I lost that edit war.
Nowadays the whole thing feels like my first "old man yells at cloud" moment, of which I'm sure I'll experience more as I age. I certainly do find new slang introduced by gen Z like "he got the riz!" to be quite cringey, so it looks like I'm well on my way to getting crotchety and terrible about the natural evolution of language! ;)
I'm broadly with you on disliking OP's comment because it is needlessly negative, but I don't think it quite rises to the standard of a personal attack.
Accusing the author of introducing slop into the language by coining (or using) a new term is a criticism of the author's work, not the author himself.
It's akin to saying, "Your work has a negative side effect which I don't like." Which clearly would've been the nicer way to say it.
The comments section on Ars is particularly depressing. I've been posting there for two decades and watched it slowly devolve from a place where thoughtful discussions happened to now just being one of the worst echo chambers on the internet, like a bad subreddit. I've made suggestions over the years in their public feedback surveys to alter their forum software to discourage mob behavior, but they don't seem to be doing anything about it.
Hello HN! I'm Eric, an experienced software developer seeking a coding or engineering management role. I've worked in many tech stacks and I'm always ready to learn new ones to get the job done. Let's make great things together!
Vanilla JS is very powerful and has the features you need to build SPAs without a big framework. Proxies and mutation observers are great for maintaining state, Updating the DOM yourself is fine, view transitions are awesome, etc. The only thing that's hard is routing, but there are lots of small dedicated JS libraries to handle that. Here's one I made that gives you the Express API on the frontend: https://github.com/rooseveltframework/single-page-express
Also personally I do not prefer to play podcasts with a podcast app. I just want it to download the files to a directory which I then sync with another audio player. Does your app make that workflow easy?
Other critiques aside, I wish there was more effort put into developing accessible CAPTCHAs that do not require JavaScript. Whatever its merits or flaws are, this CAPTCHA is yet another CAPTCHA that requires JS.
The TL;DR answer to this question is there is a lot of intellectual junk food out there and our monkey brains are pretty vulnerable to succumbing it just as we crave literal junk food and resisting those cravings is very difficult.
If only we had GLP-1 agonists for our minds too and not just our bodies.
In lieu of that, all we've got is the same as always: nurture your mind by cultivating a good media diet, a healthy skepticism that doesn't drift into reactionary contrarianism, and an openness to new information; especially new evidence that contradicts things you believe.
...Which is basically like trying to solve the obesity crisis by telling people to diet and exercise. It would be nice if we had a more effective tool or technique to help a larger percentage of people achieve it.
The machine:
CPU: 24 × AMD Ryzen 9 9900X 12-Core Processor
RAM: 128gb
GPU: NVIDIA GeForce RTX 4060 Ti 16gb (I typo'd the GPU above)
(This is via Ollama on Ubuntu.)
But 1-3 tokens per second is much faster than a lot of other high end models I've tried, so I was pretty pleased with it. Obviously other models run much faster on this hardware though.