330 karma · joined July 8, 2023
also, i'm curious, what kind of adoption were you anticipating (some time ago and now) and did the result align with it?
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?
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
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?
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)
> Unfortunately, compilers are notoriously bad at autovectorization and with the exception of relatively trivial loops, compilers rarely autovectorize effectively.
from the article
[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
(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)
always ironic when ppl say this on websites hosted in the us, a country with the most documented cases of governmental organisation backdooring/spying :/
languages don't have a strict bijective mapping between them
* futuristic buildings and flying cars *
that’s pretty cool, the vk prefix for a non-vulkan-only library is kind of confusing tho
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
america moment
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)
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 :)
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