suprjami

7 ms·

91.
by suprjami 5mo ago
Nobody does brain surgery. Nobody goes to space. Nobody is the head of state of a country. Nobody knows the airspeed velocity of an unladen swallow. Just because few specialist people do something doesn't mean "nobody" does i
92.
by suprjami 5mo ago
Certainly. You think they're going to comment? That itself would be another headline article.
93.
by suprjami 6mo ago
It also counts as overtime, right?
94.
by suprjami 6mo ago
That would be interesting. I like D2 though the lack of control over the layout is a bit frustrating sometimes.
95.
by suprjami 6mo ago
SDL has never supported DOS. You might be thinking of Allegro? https://github.com/superjamie/allegro-4.2.3.1-xc
96.
by suprjami 6mo ago
With the ease and speed of browser-based DOSBox (look at internet archive), DOS becomes a highly portable target for small games and even non-trivial games. Anything up to the complexity of mid-90s AAA titles can run practically anywhere wi
97.
by suprjami 6mo ago
Wait what. What SDL would you compile for? Win32 with exclusive fullscreen? With a VESA resolution like 640x480?
98.
by suprjami 6mo ago
The list at the top could be longer: - https://asciiflow.com/ - https://asciidraw.github.io/ Anybody know more?
99.
by suprjami 6mo ago
I already had the PC. I just mean the extra purchase of the graphics cards. The motherboard is an MSI Pro Z690-A. The slots are physical x16. Electronically they are x16, x4, x1 which doesn't harm anything at all.
100.
by suprjami 6mo ago
Actually budget friendly is RTX 3060 12Gb. With one you can run 9B/12B models which are fine for text tasks like chatting or summarisation. Not for precision like tool calling or code. With two of them you can run models up to Qwen 27B
101.
by suprjami 6mo ago
I have since found the showrunner also made the show Almost Human which I liked. I have upgraded to cautiously optimistic.
102.
by suprjami 6mo ago
If you are using quants below Q8 then get them from Unsloth or Bartowski. They are higher quality than the quants you can make yourself due to their imatrix datasets and selective quantisation of different parts of the model. For Qwen 3.5 U
103.
by suprjami 6mo ago
My favourite book finally has its TV show made. I won't get my hopes up.
104.
by suprjami 6mo ago
> created 20 days ago > multiple factual errors > advertising re-explanation Found the bot shill account.
105.
by suprjami 6mo ago
Debian build their ROCm with support for all possible devices. If you are tired of compiling from source just use a Debian Stable container, install their libraries in your container build, and pass /dev/kfd and /dev/dri
106.
by suprjami 6mo ago
If you have either PCIe slots or risers you can put them in the one system. llama.cpp will let you run inference remotely across different systems but I suspect this would be far too latent to be worthwhile. If you have three systems alread
107.
by suprjami 6mo ago
You don't. With multiple cards in normal PCI express slots, LLM layers are split across cards. When you run inference, it runs on one card then the other card. You can repeat this for as many cards as you want. You only copy the activa
108.
by suprjami 6mo ago
Some time ago I helped Jeff Olson open source his wonderful 1990s roguelike Alphaman: https://github.com/superjamie/alphaman-src It's full of optimisations to keep the code small, such as QBasic calling into C rou
109.
by suprjami 6mo ago
Just in time for Vulkan tg to be faster in almost all situations, and Vulkan pp to be faster in many situations with constant improvements on the way, making ROCm obsolete for inference.
110.
by suprjami 6mo ago
"Anthropic CEO Says AI Could Replace Software Engineers in 6 to 12 Months" https://www.entrepreneur.com/business-news/ai-ceo-says-softw...
111.
by suprjami 6mo ago
Nice to see a fellow Tornhill fan. I loved his early C articles.
112.
by suprjami 6mo ago
I would but the rest of Australia wouldn't. This country has an unhealthy drinking problem.
113.
by suprjami 7mo ago
Yes, AMD themselves even use Vulkan tg numbers in their marketing material, because it's faster than ROCm on everything RDNA2 onwards (seems embarrassing). However for pp, Vulkan is still nowhere near close to ROCm. That matters for lo
114.
by suprjami 7mo ago
Following the current rule of thumb MoE = `sqrt(param*active)` a 200B-A3B would have the intelligence of a ~24B dense model. That seems pointless. You can achieve that with a single 24G graphics card already. I wonder if it would even hold
115.
by suprjami 7mo ago
I'd strongly support a year-based ban on cigarette purchases. Set the purchase birth year to the current age 18. So DOB 2008 if done today, if you're born 2009 or later you can't buy smokes at all ever. Within two generations
116.
by suprjami 7mo ago
You have got to be joking. My three NVIDIA cards are more power efficient than my one AMD card, both at idle and during usage. Official ROCm is like pulling teeth with poor support for desktop cards. Debian, a volunteer led project, have be
117.
by suprjami 7mo ago
Some models really suffer badly from KV quantisation. You can also take a speed hit using dissimilar K and V types. TurboQuant seems to be the next big thing in context memory usage. Polar coordinates achieving ~5x reduction in memory usage
118.
by suprjami 7mo ago
Dual 3060s run 24B Q6 and 32B Q4 at ~15 tok/sec. That's fast enough to be usable. Add a third one and you can run Qwen 3.5 27B Q6 with 128k ctx. For less than the price of a 3090.
119.
by suprjami 7mo ago
Look up "USB RF remote" on eBay. There are two common ones you'll see everywhere. I have one for my Kodi system.
120.
by suprjami 7mo ago
Please anyone make a new Slack. 4Gb RAM for a slow chat client with a bad interface is just so slovenly it should be illegal.

More ›