Thanks for the clear instructions. When I installed and tried running the cuda samples, it errored out. It got fixed when I installed cmake. sudo apt install cmake
Your personal story is so Epic that you redefined my course in life. I taught it was impossible to reach Big Tech without Ph.D. and you are the very living proof that it is. Thank you so much for your existence brother, now i believe that i can do that too.
Hey, sorry, 21st minute, you say we need r+1 layers for information to have propagated to a node r-hops away. Why is that? Shouldn't r layers be enough? For example, a 1-layer MPNN will aggregate information from B_1 (nodes 0-hops and 1-hop away).
Pretty sure cuDNN installation on linux has made people self-delete before... They really do make it harder than it needs to be. If anything updates you gotta do it over too.
In a year, it's going to look really stupid everyone in 2024 paying large chunks of money for video-game video cards so that they can do AI/ML. For one thing, none of these cards have the RAM to run even modestly sized models.
So what do i do? How does anyone manages to run AI and Machine Learning if modest size models cannot be runned? Should i buy a pc and build it or there is a good laptop for it? Please enlighten me.
@@JeremyWinston-fo5fz Seems like the answer is a lot of RAM with an NPU. It runs LLMs faster than a GPU and you are not bottlenecked with the amount of RAM on the GPU card, which is usually not enough to hold a full model. Even just a fast CPU would be fine, like a 7950x or an ARM chip (M4). But even a cheap NPU addition can get things like a Raspberry Pi running LLMs. It seems incorrect to spend $1k+ on a GPU with only 12GB of RAM with the intention of running a 70+GB LLM.
Wonderful video, to the point, starting with the paper, understanding the necessary background and moving forward with practical exercise, very nice, keep up the good work bro.
So, let's say we have trained all animals with images and texts except wolf and dog images and texts, if we were to ask to a model which has been trained with CLIP and zero-shot classification to draw a picture of a dog and a wolf sitting together, would it still be able to draw them? But it hasn't been trained neither the words of dog and wolf nor images of wolf and dog, if we were to describe as "draw a picture of an animal which howls and looks similar to a fox or canine and lives in the forest" there is a chance it would be able to predict it, but if we say the exact sentence of "draw a picture of wolf" it wouldn't be able to draw it, right? Because it doesn't know the meaning of "wolf", or how it looks like, or it is an animal or a table, in its universe it would be like humans trying to predict how 10th dimension looks like? Am I right?
Thomas joined me to discuss LLaMA 3 work! Also - if you need some GPUs check out Hyperstack: console.hyperstack.cloud/?Influencers&Aleksa+Gordi%C4%87 who are sponsoring this video! :)
Great video. I'm starting my research for machine learning. I subscribed after looking at the video. Hopefully you do the same for specific needs. I'm interested in trading.
Various machine learning experts - real experts who design and run ML systems as a business, not amateurs on RU-vid - recommend 4 CPU threads per GPU. Considering a dual GPU system, the Ryzen 3600 is the cheapest 12 thread CPU. Alternatives from newer generations are Ryzen 5600, Intel 12400F and Ryzen 7600. All Intel CPUs consume more than double the amount of power per performance than AMD CPUs. From a long term cost perspective, chosing AMD is a no brainer because the Intel CPUs will cost a fortune on electricity bills.
Source? I do not doubt you but i am struggling, i don't know what to buy to make my first pc for deep learning, or if buying a laptop would be better. My budget is 1500 more or less.
23:00 Interesting take, but I feel like this is a "greedy search" approach to technological development, when our society could really benefit from look-ahead. I would have been interested in what he would have said if you pressured him to talk about what he thinks beneficial *pure* research, not translating quickly into products, would look like.
We had Junyang author of Qwen 2 one of the best LLMs with us today! If you need some GPUs check out Hyperstack: www.hyperstack.cloud/free-credit-landing-page-op-2? who are sponsoring this video! :) and you'll also get some free credits ;)
Thanks a lot for your videos on GNNs! After scrolling through multiple papers/articles and not understanding them, I am finally starting to get some intuition on GNNs after watching your videos. And now going back to those papers, I started to actually understand them. Thanks a lot!
The adversarial loss - i think the explanation is wrong You said the discriminator tries to maximize it, however, you have just shown that it tries to minimize is (the term becomes 0 if D(x) is 1 and D(\hatX) is 0). So the discriminator tries to minimize it (and because its a loss function it makes sense), and the generator tries to do the opposite, maximize it, to fool the discriminator. So I think you mis-labeled the objective: L_GAN we try to minimize (minimize loss) in order to train the discriminator.
Good coverage of the paper. I like how you go into each aspect in depth. I think you struggle--and you acknowledge it yourself--explaining some of the more complicated parts of the paper. If you can nail how to communicate those you would make an awesome teacher.
Hej, video sam da si na svojoj stranici napisao da si zavrsio etf, pa me zanima da li si pod onim opisom mislio na to da si zavrsio smer za elektroniku? 😄 Pozdrav iz Srbije!
The loss part is still very confusing to me, why add the same loss twice with different weights, and as you said in the end going to zero? Anyone knows?