HNHacker News
TopNewBestAskShowJobs

taminka

330 karma · joined July 8, 2023

submissionscomments
taminka··on Ollama now supports AMD graphics cards
why though? except for apple, most vendors still actively support it and newer versions of OpenCL are released…
taminka··on Radicle: Open-Source, Peer-to-Peer, GitHub Alternative
i really like the website and application design, bc so many oss projects often completely falter w/ visual design, and while this is a superficial thing, beautiful design makes me want to interact w/ a project more :)

also, i'm curious, what kind of adoption were you anticipating (some time ago and now) and did the result align with it?

taminka··on GGUF, the Long Way Around
wait, why is embedding the graph into the file bad?

it enables really clean separation of the core autodiff library and whatever backend you want to use to accelerate the graph computations, which can simply read the file and be completely independent of the core implementation

but also, if you just store the tensors in some arbitrary order and then store the indices of the order in which they have to read and traversed, you can easily adjust the graph to add stuff like layer fusion or smth similar (i'm not really familiar w/ comp graph optimisations tbh)

what would an alternative look like anyway?

taminka··on Optimize sgemm on RISC-V platform
wait, are you suggesting apple has some hardware level GEMM instructions that are undocumented or smth? also, what do you mean by system-provided BLAS? their accelerate framework?
taminka··on Optimize sgemm on RISC-V platform
yeah, that's probably it

i'm using an m1 laptop, so my sgemm implementation relies on arm neon vector extension, and i saw literally no performance improvements when i added prefetching, which could either be a result of their speculative optimisations or an issue in my own implementation

but i have noticed some ppl saying that adding prefetching on x86 has little performance improvement for sgemm, bc it's a scenario in which it's easier to predict the next memory address that's gonna be used

taminka··on Optimize sgemm on RISC-V platform
interesting write-up even outside of the risc-v context, bc most of the optimisations are the same for different platforms

one interesting difference is that memory prefetching/preloading has such a huge difference, whereas on arm/x86 it barely makes any difference (in my experience) that it's not even worth doing

also, doesn't allwinner d1-h have simd registers?

taminka··on Cake – C23 and Beyond (2023)
lowkey smart pointers are often just used to deflect the responsibility of thinking about memory layouts

https://floooh.github.io/2018/06/17/handles-vs-pointers.html

and most issues can be caught by using a static analyser of a memory leak checker (getting ppl to consistently use them is another issue, but still)

taminka··on We Have to Start Over: From Atom to Zed
i'm not saying that manually writing simd assembly or intrinsics is useless, it's very often necessary, i'm just disagreeing w/ the statement

> Unfortunately, compilers are notoriously bad at autovectorization and with the exception of relatively trivial loops, compilers rarely autovectorize effectively.

from the article

taminka··on We Have to Start Over: From Atom to Zed
despite what the article claims, clang and gcc are acc very good at vectorising code, and intel, arm, etc have entire teams dedicated to improving vectorisation in modern compilers
taminka··on Conformant OpenGL 4.6 on the M1
> I've read that generally opengl is just easier to use than vulkan

[here's](https://learnopengl.com/code_viewer_gh.php?code=src/1.gettin...) an opengl triangle rendering example code (~200 LOC)

[here's](https://vulkan-tutorial.com/code/17_swap_chain_recreation.cp...) a vulkan triangle rendering example code (~1000 LOC)

ye it's fair to say opengl is a bit easier to use ijbol

taminka··on Understanding Automatic Differentiation in 30 lines of Python
by "numerical" i meant "operating on numbers as opposed to symbolic values", not "using the method of finite differences", apologies for the confusion
taminka··on Understanding Automatic Differentiation in 30 lines of Python
chain rule is defined for partial derivatives, so it's still technically just chain rule
taminka··on Understanding Automatic Differentiation in 30 lines of Python
everyone studies chain rule in school, so it's definitely more clear, especially if you look at the amount of unnecessary complexity and convolutedness in many autodiff explanations (an substantial amount for such a simple subject)
taminka··on Understanding Automatic Differentiation in 30 lines of Python
i wish people would just call autodiff (or at least describe it) as numerical chain rule, because that’s quite literally all it is (plus a few tricks for not explicitly computing the jacobian for certain operations), it’s a lot more clear that way
taminka··on Hugging Face raises $235M from investors including Salesforce and Nvidia
i'm curious what HF does, because iiuc they're mostly just hosting model files? i know that they have some compute offers, and also created/maintain a few libraries, but it's not particularly widely used, i'm really not sure how they're supposed to earn money...
taminka··on PDF Tool – Modify PDFs in the browser without uploading
i'm even less virtuous than russia or china, yet i can't spy on you

(once again, it's a combination of current technical capability, concentration of current and early important technologies being developed in your jurisdiction, concentration of current and popular technologies being developed in your jurisdiction, etc)

taminka··on PDF Tool – Modify PDFs in the browser without uploading
backdooring and censorship are two very different things
taminka··on PDF Tool – Modify PDFs in the browser without uploading
> going to Russia? or China?

always ironic when ppl say this on websites hosted in the us, a country with the most documented cases of governmental organisation backdooring/spying :/

taminka··on Has Google Translate been fixed yet?
maybe the word is a homonym in one language, but isn't in the other

languages don't have a strict bijective mapping between them

taminka··on Candle: Torch Replacement in Rust
what being a kitchen sink language does to a mf
taminka··on VkFFT: Vulkan/CUDA/Hip/OpenCL/Level Zero/Metal Fast Fourier Transform Library
* world if everyone just used vulkan *

* futuristic buildings and flying cars *

that’s pretty cool, the vk prefix for a non-vulkan-only library is kind of confusing tho

taminka··on IronOS: Open-source soldering iron firmware
it’s not like it connects to internet or anything, having a digital display to set power/temperature is pretty convenient
taminka··on Show HN: LearnLingo – Converse with an AI-powered language tutor
id say it depends

i learned english to a significant degree by simply conversing with ppl on the internet and watching tv shows/movies in english

for me something like this would be tremendously useful

taminka··on Show HN: San Francisco Compute – 512 H100s at <$2/hr for research and startups
ask no questions hear no lies
taminka··on An ultra-sensitive on-off switch helps axolotls regrow limbs
> Imagine if we can get to a world where the difference between not having health insurance and having insurance means either you wait for your body to naturally regrow a limb or your insurance covers a more quickly lab grown one.

america moment

taminka··on Stable Diffusion XL 1.0
https://huggingface.co/stabilityai/stable-diffusion-xl-base-... :)
taminka··on Sugary drinks tax could prevent decay and increase health equity: study
i mean, for example here in europe there’s no such iiuc, sugary drinks are still way too sugary and way too cheap
taminka··on Llama2.c: Inference llama 2 in one file of pure C
not really imo

i'm really enjoy the resurgence of very minimal implementations of ml algorithms, because if you've recently tried performing inference on a sophisticated ml model in a way that's user friendly in any capacity, you know that it essentially involves pulling out your prayer book, rosary and incense, pulling like 20gb of python dependencies, 20 different frameworks, all of which breaks very easily, any minor difference in versioning is guaranteed to break the entire setup, with no hope of fixing it, it's just bindings on top of bindings on top of bindings, every other day a new library comes out that builds on top of existing libraries, introducing their new format, promising "deploy models in with 15 lines of python", then "10 lines of python", then "1 one of python", which essentially calls into a black box N layers of python on top of each other, calling into an extremely complicated C++ autodiff library, the source code of which can only be acquired by an in person meeting with some sketchy software engineer from czechia, all of which only works on python 3.10.2, cuda v12.78.1298.777 with commit aohfyoawhftyaowhftuawot, only compiled with microsoft's implementation of C++ compiler, with 10 non-standard extensions enabled, all of this OF COURSE only if you have the most optimal hardware

point is, if your implementation is a simple C project that's trivial to build/integrate into your project, it's significantly easier to use on any hardware, not just retro (popularity of llama.cpp is a great testament to that imo)

taminka··on Toyota has been developing a solid-state battery for EVs with a range of 745mi
> For those who prefer metric, that’s a range of 1200 kilometers and a charge time of six hectoseconds

my brother in christ, literally the entire world not only prefers metric, but it’s the only one we know

like, i don’t mind seeing miles, esp given that it’s a US publication, and i already pretty much automatically convert from miles to km in my head, but please be humble with your silly system :)

taminka··on Anytype – local-first, P2P Notion alternative
real

a libp2p based wysiwyg or even a plaintext app with p2p sync + works offline + encrypted + cross device sync + collaboration with other users would be invaluable

← PreviousPage 5 of 6Next →