How to Train Your Models in the Cloud
Вставка
- Опубліковано 11 тра 2017
- Let's discuss whether you should train your models locally or in the cloud. I'll go through several dedicated GPU options, then compare three cloud options; AWS, Google Cloud, and FloydHub. I was not endorsed by anyone for this.
Code for this video:
github.com/floydhub/fast-styl...
Please Subscribe! And like. And comment. That's what keeps me going.
High Budget GPU: Titan XP www.amazon.com/NVIDIA-GeForce...
Medium Budget GPU: www.amazon.com/MSI-GAMING-GTX...
Small Budget GPU: www.amazon.com/dp/B01MF7EQJZ
Build a Deep Learning machine:
/ build-a-deep-learning-...
/ building-your-own-deep...
www.oreilly.com/learning/buil...
More learning resources:
www.infoworld.com/article/3179...
thehftguy.com/2016/06/15/gce-...
/ aws-vs-google-cloud-fl...
news.ycombinator.com/item?id=...
Join us in the Wizards Slack channel:
wizards.herokuapp.com/
And please support me on Patreon:
www.patreon.com/user?u=3191693
Follow me:
Twitter: / sirajraval
Facebook: / sirajology Instagram: / sirajraval
Signup for my newsletter for exciting updates in the field of AI:
goo.gl/FZzJ5w
Hit the Join button above to sign up to become a member of my channel for access to exclusive content! Join my AI community: chatgptschool.io/ Sign up for my AI Sports betting Bot, WagerGPT! (500 spots available):
www.wagergpt.co
Siraj, congratulations on your work. It's amazing. Shout out from Brazil!
Lol Linus with a GPU was the best part xD
Oh boy, it was damn funny..
timecode please
1:33
thank you, i have actually been researching the best way to do this since the gru video. thank you so much
Huge shot out! I have 5 days to train this model for my DL class and I was getting so many problems to set up training on the cloud
Your background for the first ten seconds is 🔥
Thank you very much Siraj, very good & interesting content as always! :)
I'm just getting started with DL so i will use my dell xps 15 notebook with gtx 1050 graphics for the start, but as soon as i feel i need it i will definitely give floydhub a try.
I'm a student (Communications and Computer Engineering) and don't have the money to buy a big expensive local desktop (and i prefer portability)
Fuckin' LOVE this channel, my man. Glad I found you, learning a ton.
Thanks A Lot Siraj!!!! This video was really helpful!!!
Hey SiraJ, is there a way to aggregate 2-3 GPU cards for a certain ml training task?
Awesome Siraj... very informative..
So much supportive video, thanks Sir
This is really great! Experimenting with FloydHub as we speak.
that was suuuper easy! thanks siraj
I did my whole Udacity DL nano degree using AWS... I loved it... the trick was to terminate the instances after use so you don't accrue higher charges for storage and stuff..
awesome
Yeah I learnt that the hard way unfortunately...
Yep also made that mistake :(
got stung with that one! £300 worth. Moral of the story: Turn off AWS instances when you aren't using them!
Amazing like always !
thank you for deep-learning cloud video
hey @sarajraval , can u suggest some good AI/ML project for my final year engineering project? really having a hard time getting good ideas since I've no idea of this domain. big fan btw !! u make it look so easy.
how much faster was it to run in the cloud vs. regular spec on your laptop?
Thank you Siraj, that helps me a lot.
你有没有试过阿里云,它好像提供M40显卡,一直到9月份都免费
I saw you recommended Nvidia gpus, but could a radeon gpu also work? Vega has pretty high FLOPS. Do you need to use cuda for everything, or does opencl also work?
Great video as always! Many thanks for your amazing work!!!!!
Just one feedback, your channel is amazing and full of resources but there is no way to know in which order to go through the contents. Naming videos with order and/or making a mind map or whatever to order your contents would an addition of great value ;-)
(I know the Udacity content is ordered)
dude tnq so much for the info
Cool!! I've been using AWS to practice, but it's a little pricey. I'm probably gonna use Google Cloud + Floydhub until I've used all my credits.
very nice . good job , all people like it , good luck , thanks,
I prefered google cloud so far, however i will try out Floyd to practice training. Thakns for the video
Can I run a hello world script on gpu.?
I mean, are only some frameworks like tf allowed to be run on gpu?
So if I wanna start in deep learning is the best to buy a laptop with a GTX 1060 or a portable computer and then use floyd ? Help !!
First (17 Views), I asked my prof: Whats the diffrence in the calculation between cpu and gpu. He answered, GPU are better on float calculation. But your explanation is much better.
Did you try Google Cloud ML? It seems to accept tensorflow model and dataset on input and train it cost-effectively.
Siraj, love your videos congrats. Could you please do one for deploying a web ML app into production on AWS?
So either way, if I choose Google or Amazon they will be able to see my developments and practices? And the data we use is it free, it's clean data, or what?
hey Siraj,
do you still have the same recommendation now that google has announced their new cloud architecture?
Thanks a lot Siraj
can you do a vid IBM Watson vs Tensorflow for stock predications and using the cloud to scale the computing power
I love your video ,,,You are amazing.
thanks a ton!!! didn't even know for floydhub
np
Hi siraj.. will you say me how to deploy deep learning in mobile ?? I got confused of choosing gpu ..my theme is about prediction of cancer by images in mobile using deep learning. I need configuration it makes me to have clear picture .pls help me
Hey siraj, can you tell difference between a cell and a neuron wrt lstm net ?
Is there any cloud GPU to run deep learning code in "JAVA "
Also colab research offers a tesla k80 for free. Is it a good option?
Hi Siraj can you make a video or recommend a video on hyperparameters optimization? Thanks
Dude you're awesome! Please do more, and keep using memes.
thanks again, Siraj.
Hey Siraj! Will you be doing a video on converting 2D images to 3D models using deep learning? I would love to see that.
autodesk remake uses it, its really cool
You are brilliant and so are the memes and graphics/videos; who does those?
thx its all me
it is no longer the titan xp, it is the tesla V100 with 120 teraflops of deep learning performance :).
What is the defference between those clouds and google colab?
I wonder: 1) Is cloud computing on Google's GCP automatically parallelized (means exploiting all given GPUs and CPUs optimally)? 2) Can I use one GCP account from several computers? (We need it at my company and several coworkers have to use the service)
I got exposed to this big data stuff on your channel , and i just found another reason to love computers :D !!
please make more detail videos on this topic. Like how to train in google cloud and aws platform
How to make a Deep Learning Program that learn to Delete the Blur of an image ?
Thanks for letting us know about Floyd. It's straightforward to use, I think.
What about ML cloud from google? How are those instances configured?
THANK YOU!
how can you run a carla simulator on the cloud
Please review the Titan V. What's the real life speed up?
is Gt840M Better For Deep Learning?
What about ibm bluemix cloud platform?
why not use a asic card for deep learning? how is asic card for bitcoin mining different from these GPU's ?
Could you cover IBM BlueMix?
Can u do a video on how to setup Google cloud for deep learning?
What is the differences between Google Colab and Google Cloud?
What about TPU Cloud?
hi siraj,
views on azure?
siraj any special video to be released for patreons?
thanks
Great vid bro! But can anyone clear my dumb? Should I use cloud computing when learning deep learning subject?
This is the speed you should go at. Slow enough to understand, fast enough to not get boring.
How long will it take to train a model like ResNet using GTX 1050 ? Let's say I will use it for typical kaggle competition. Is it enough to build a complex model that will lead me to being in high rank?
I'm using a pretrained resnet on my project, I have a gtx 1060 3gb, and believe me it's a hassle. The model barely fits on the memory, then I had to resize all the input images to 150*150 so that they fit. Typically in deep learning you'd want at least 4 gb. Many people recommend the 1070 and above.
Thank you!
Hey guys! I was wondering what are the economic laptop options for deep learning. Also, please give me suggestions which are relevant for India.
By the way, which one is Pink?
Hi developers , what do you think about Paperspace ?
suposedly keras works well with amd cards , so it is best to use it or theano , to take advantage of cheaper hardware
was eating chips and could not hear half the video plz fix
I think on AWS a spot instance with GPU (tesla k80) is around 0,20$ while a the same preemptible one on google cloud is 0,45$. Of course the pay per minute is a advantage.
Using this cloud can I able to run yolov3
How to run a model on cloud and test it into android?
Hey Siraj, can you make a video on how to use Google Cloud ML Engine?
How about paper space ??
Harold & Kumar is the name of a series of American stoner comedy films starring John Cho and Kal Penn. The first film, Harold & Kumar Go to White Castle, was released in 2004 by New Line Cinema and was followed by two sequels, Harold & Kumar Escape from Guantanamo Bay and A Very Harold & Kumar 3D Christmas.
Jetson TX1 is half price currently, and an awesome platform
But Tesla k80 is now also used by Google colab like the last platform you talked in this video, though colab provides unlimited usages of their gpus. So it's better then heroku for deep learning providing only 100 gpu usage hours
Hey Siraj, thank you for your objective view on which cloud provider to use. Can you please tell us what is your favorite option to go with?
i hear that upcoming boards of the new risc v cpu architecture will be relatively cheap , but good for training as they will have a massive number of tiny cores with support for half precision logic
and by hearing i mean thinking
pls show more of the results!!! like pictures
AWS now offers per-second billing on their instances
3:01 That's good.
Is the floyd hub free?
Just to train a model, you don't need well known services like aws and google, bc they are for big projects. For one task a small cloud provider and that will be seversl times cheaper
I bought a Lenovo Laptop with a Quadro M1200 lets hope its good for DL. Now to just learn how to program well! xD
Please make a video of Google's TPUs
Google Cloud is not cheaper than AWS, a lot of the time the biggest factor in the price for cloud computing is data transfer out. On AWS it's $0.07, on Google Cloud, it's $0.12 per GB. If you compare a lot of the services that are similar, Google is more expensive, for example, Google Cloud Functions are $0.40 per million invocations whereas AWS Lambda is $0.20 per million, and the execution time is more expensive on Google.
really you made it for programmers .... even kids will get bored of this
Update: AWS is also available in pay per minute model now.
What about H20.ai & algorithmia
Would u plz, add a video describing google colab plz
what about MS Azure ?
similar to AWS
I compared the new Azure vs. AWS vs IBM bluemix, with massive networks (ResNet) - Azure was the best, both in performance and price.
yeah, I'm using Azure . Thanks for clearing.
What about not MS azure?
nah
can i use this to train an different painting image and apply it on another image?
1000th like!
You forgot to mention Nvidia's new Tesla Volta!
Also, It's worth mentioning that GCP might make TPUs available on their platform. That'd be amazing.
Correction: they just announced TPUs will be available for GCP at this year's I/O conference!