DiabloD3

31 ms·

421.
by DiabloD3 1y ago
This is spam.
422.
by DiabloD3 1y ago
This isn't a new feature, and was announced for... I think 8. Why is it being re-announced now?
423.
by DiabloD3 1y ago
llama.cpp is the SOTA inference engine that everyone in the know uses, and has a Vulkan backend. Most software in the world is Vulkan, not CUDA, and CUDA only works on a minority of hardware. Not only that, AMD has a compatibility layer for
424.
by DiabloD3 1y ago
Man, best /s text I've seen on here in awhile. I hope other people appreciate it.
425.
by DiabloD3 1y ago
I don't bother with Nvidia products anymore. In a lot of ways, they're too little too late. Nvidia products generally perform worse per dollar, perform worse per watt. In a single GPU situation, my 7900XTX has gotten me farther th
426.
by DiabloD3 1y ago
It isn't odd at all. Ollama uses an ancient version of llama.cpp, and was originally meant to just be a GUI frontend. They forked, and then never resynchronized... and now lack the willpower and technical skill to achieve that. Ollama
427.
by DiabloD3 1y ago
So where does the magnetic field come from? ;) ;) ;)
428.
by DiabloD3 1y ago
They need to reverse course on this immediately, along with removing all the other LLM-based technologies. No mentions of "AI" should appear anywhere in the OS. According to the article, Microsoft will be able to read all of your
429.
by DiabloD3 1y ago
Yes, we know. It is also brighter the further you die from home.
430.
by DiabloD3 1y ago
In other words, completely true because this is HN and we know things that the other 99.999% of the world doesn't and never will and never even thinks to ask . Instead, "Google took away my favorite software off my device that
431.
by DiabloD3 1y ago
"If you find a better deal, shove it up your ugly ass. You heard us right! Shove it up your ugly ass."
432.
by DiabloD3 2y ago
Even saying 25-30GB/s is weird. DDR4-3200 is ~26GB/s per channel, and is the upper end of what you'll see on ECC DDR4. DDR5-5600 is common now, and is ~45GB/s. Zen 2/3 Epycs on SP3 have 8 channels, Zen 4/5 Epyc
433.
by DiabloD3 2y ago
We already don't need a crude process like you describe for lighting applications: look up "chip on board" LEDs, or COBs. The brightest single element LED lights you can buy today are all COBs. What we need is finer COBs that
434.
by DiabloD3 2y ago
"Gaia uses the open-source Lemonade SDK from ONNX TurnkeyML for LLM inference." It says it right in the article. Its all Python academic-quality code, and I don't know why AMD isn't partnering with high performance infer
435.
by DiabloD3 2y ago
Neat idea, but why is this implemented in Python instead of a more suitable language? Also, there are many DSLs and APIs already for this, some even supported by industries and provided as part of the SDKs for various hardware blobs, so why
436.
by DiabloD3 2y ago
Depends how you define "worse in every way". With my own personal testing, DeepSeek's distillations have been able to do the task when the original upstream model either couldn't, or was marginally worse yet. You're
437.
by DiabloD3 2y ago
No one said they were. They're distilled using the original model and the same weights that match the ones in R1. Its ostensibly the original, but better. There are also fused models such as https://huggingface.co/FuseA
438.
by DiabloD3 2y ago
But you just described what Mozilla is doing right now! 0% of what anyone donates goes to the browser, and its a disaster!
439.
by DiabloD3 2y ago
Yes, I'm aware of various rankings. Try all of those models on something that isn't commonly used on a benchmark, and you'll notice that a lot of the proprietary models have trouble actually producing statistically relevant r
440.
by DiabloD3 2y ago
You wouldn't use full fledged R1 for coding. There are distilled models using R1 for coding that get you most of the way there. R1 also doesn't take 1TB of RAM, go use read Unsloth's writeup on how to reduce model size withou
441.
by DiabloD3 2y ago
I'm sorry, but isn't Cursor just an editor? Maybe an editor shouldn't actually have garbage parts to avoid? Why not just use an editor that is focused on coding, and then just not use an LLM at all? Less fighting the toolin
442.
by DiabloD3 2y ago
The only way I would donate to Mozilla is if the corporation is shuttered and the non-profit is disentangled from it. Any donations you send to Mozilla today go to the corporation and are not spent on the browser. They are spent on things t
443.
by DiabloD3 2y ago
Yet many Android phones had the "notch" the same generation as Apple did, and it was smaller; Android phones also quickly eliminated the notch and went to a small hole, and now phones have hidden cameras entirely. Apple needs to
444.
by DiabloD3 2y ago
Ollama's library butchers names, I believe its this: https://ollama.com/library/qwq The actual name (via HF): https://huggingface.co/Qwen/QwQ-32B
445.
by DiabloD3 2y ago
So, I'm usually the huge Rust promoter, but we already have a lot of dotfiles managers, and Chezmoi, although written in Go, is right there. What can this one do that the other ones don't? The README lists features that, well, I&#
446.
by DiabloD3 2y ago
Yep. And they don't even have skin in the Windows game. They haven't been the Windows company for... Oh Christ, I'm getting old, since 2014. The day Sataya became CEO they officially became a cloud company. If it wouldn'
447.
by DiabloD3 2y ago
I'm going to be honest, but this is a really weird way for Microsoft to announce Edge is EOL and they can't afford to even hire two or three more developers to perma-fork Chrome and bring the rest of the Chromium community under o
448.
by DiabloD3 2y ago
With assembly for a CPU, the instructions you're writing are verbatim being sent (compiled, obviously) to the CPU. With PTX, you are not writing in the native assembly of the Nvidia GPU, but yet another abstraction, just one more simil
449.
by DiabloD3 2y ago
"Compile to" isn't exactly the correct phrase either. PTX is not the IL used by Nvidia's drivers, but does compile directly to it with less slop involved. If you had said "PTX's instructions are analogous to wr
450.
by DiabloD3 2y ago
There isn't a lot of easily accessible examples outside of the corporate world. Open source authors typically shy away from Nvidia closed source APIs, and PTX is tied to how Nvidia hardware works, so you won't see it implemented f

More ›