machinekob

20 ms·

181.
by machinekob 4y ago
Im more about "East Asia is the only region which was certainly 'poor' but now can be described as somewhat 'rich',"
182.
by machinekob 4y ago
East/Central Europe (post commie)?
183.
by machinekob 4y ago
Ok show me how to use Rust for my big deep learning projects or even in big iOS projects and i'll switch :D
184.
by machinekob 4y ago
Citing other comment -> "Tachyum is writing press releases like a machine gun (2-3 a month), but they don't participate in MLPerf and avoid comparisons altogether. They talk about big facilities, but they have 43 employees (in
185.
by machinekob 4y ago
But Intel is investing in ARM/RISC-V and x86 will be with us a bit longer (probably even 10+years). I would be more concerned about AMD then I'm about Intel tbh.
186.
by machinekob 4y ago
It looks more like B2B cloud at least from past informations about it not sure about AWS rivality.
187.
by machinekob 4y ago
Nice post but I saw it few days ago on hacker news or it is deja vu?
188.
by machinekob 4y ago
Not even one is close to colab yet (first of google TPU is only way to access GPT-like models for most people, second you can get pretty neat GPU for few dollars when on other platforms you are paying a ton of money for that)
189.
by machinekob 4y ago
I think we are lock out cause of gigant-scalers (cloud providers) and labs maybe in few years we can get some cool ARM servers for cheap :P.
190.
by machinekob 4y ago
Make people depends on ecosystem. I'll see AWS/Azure/NVIDIA doing smth like that as that just to create they own ecosystem as its extremely profitable to sell gpu VM's for them.
191.
by machinekob 4y ago
Good for all of us maybe some good alternatives to colab will be created cause of that.
192.
by machinekob 4y ago
Ahh nvm I was thinking about m1 max (my brain is damaged by the m1 naming and i didnt saw information that you are using m1 pro gpu)
193.
by machinekob 4y ago
Weeks :? Looking at every other project from Apple that try to integrate with some opensource lib it'll be months/years if they didn't drop support for it before that :P Did you get only 2.55x on BERT/other transformer
194.
by machinekob 4y ago
MPS (Metal backend for PyTorch) is pretty poor integrated right now most of my custom models are crashing and pure Swift + MPSGraph versions is working 3-10x faster then PyTorch. So I'm pretty sure there is A LOT of optimizing and bug
195.
by machinekob 4y ago
Good researcher != good director/team leader.
196.
by machinekob 4y ago
I can assure you that more people exist with B1 knowledge of english then folks that have at least minimum knowledge of all math and computer science related domains that are used for deep learning (even subset like audio/video) even i
197.
by machinekob 4y ago
I strongly agree with readability in my opinion its cause Academia people live in "bubbles" and they assume everyone knew what a domain specific terms and greek letters means so its easier to read some omega then for example learn
198.
by machinekob 4y ago
Sadly its dead project
199.
by machinekob 4y ago
Biggest problem for diffusion models were performance (as you need to iterate even at inference) But I'm not up to date with newest architectures maybe its already solved :P
200.
by machinekob 4y ago
Probably cause all of the other cloud providers are limiting operations in Russia
201.
by machinekob 4y ago
So question from novice in containers, podman vs docker why should i consider switching from docker?
202.
by machinekob 4y ago
You don't need juliaconda with julia cause default package manager just works good enough. But i can agree community is way 2small and compared to big languages tooling are extremely poor.
203.
by machinekob 4y ago
Cool that tiny models have better performance on some Julia frameworks then on pytorch but till Julia get some framework that allows for seamless multi-gpu training a lot of DL people just wont use it sadly. Its for sure easier to just opti
204.
by machinekob 4y ago
As it is in China you can bet on it. There were no major china tech company that product something and do not steal tech from "west" yet (maybe there are smaller companies but no big tech got it yet).
205.
by machinekob 4y ago
I would say https://github.com/arogozhnikov/einops should be pretty close to your requirments
206.
by machinekob 5y ago
What about inference latency? Im only aware of audio diffusions model but even there latency was pretty big (2/3 years ago)
207.
by machinekob 5y ago
Mb im blind ofc Cython would be faster :P
208.
by machinekob 5y ago
Would love to but community is way to small for my use case :(
209.
by machinekob 5y ago
Python with type annotations isnt faster using python runtime. But some packages can utilize it for higher performance but most of the time it'll be slower cause you need to parse extra information if you want to reuse it in pure pytho