Timisoara study group - fast.ai Live

We’ll be organizing weekly group studies in Timisoara, Romania. Anyone from TM or around is more than welcome to join. Details still in the works but will keep everyone posted here and on the local deep learning meetup group :
Timisoara Deep Learning Meetup http://meetu.ps/c/32gXJ/zC2Xb/d

2 Likes

I live in Arad and work a lot, so it is not easy for me, but if we organize it late afternoon or evening, during the weekend, it will be ok. Eventually I can try to organize another group of around 10 friends in Arad to work with and we will relate to this Timisoara study group,

Most of us work and/or study so we will surely meet during the evenings or on weekends ! It will be great for the Timisoara and Arad deep learning communities to collaborate and share knowledge.

As promised, I’ve compiled a short comparison between best free communication tools, we may use:

Slack
+ probably easiest
+ most 3rd-party integrations – don’t know how much this would help/affect us
- 10.000 message history limit – easier to hit than it seems
- no group calls/video chats

Discord
+ very fast/snappy
+ no message limit
+ no total space limit
- max 8 MB file upload size
- no useful integrations (although workarounds exists)

Microsoft Teams
+ no message limit
+ 10GB total space limit (Slack: 5GB)
+ video conferences for 80 people
-/+ less integrations, but more free ones
- harder to manage – only read, haven’t tried out (yet)

1 Like

Here’s the rest of the summary from our meetup:

  1. We’ll start with one study group every week, on Friday evenings (I propose 7 PM, starting on October 26th), so it’s easier for our colleagues from Arad to join us. Also if we do long coding nights well … no work/school on Saturday
  2. We still need to find a place to meet for the group study sessions but for that we need to know how many of use there are first. So everyone please send me a private message so I know you’re here and we don’t spam this thread. I’ll also use the usernames to set up a private message group for us.
  3. Courses will be at 4:30AM EEST so most people won’t be able to attend them live but I’m 99% sure they will be available in youtube (unlisted)
  4. You’ll need either a decent Nvidia GPU (at least 9xx with 4GB RAM) or a cloud GPU instance, see : https://github.com/binga/cloud-gpus ; alternatively, you’ll could negotiate with @3DSF_LTD access to his GPU server :slight_smile:

– Do you mean also next Friday 12 October, or do you mean the first Friday after the first lesson?
– For the GPU’s whoever wants to work on my laptop is welcome during the study group of Friday. I know also other guys in our group has a GPU laptop, so it should not be difficult to organize there.
For the GPU servers I have in Arad it is a little bit complex for the moment from Timisoara. But whoever wants to come in my office can work on my servers locally.
WELCOME!

Possibly Virgil can also evaluate on his experience, for people with laptop WITHOUT GPU, if it is possible for them to work on their and connect to the available GPU laptops to run the code when needed.

after first lesson, i’ve edited the original posts.

I use teamviewer & anydesk to connect to my desktop, so if someone has a desktop with GPU and laptop without one, this should work for them. But this approach works best if the server is not shared.

Could be better to connect via ssh to a server with GPU but I haven’t tried it. If anyone has more experience with this please share.

I do have GPU on my laptops but only AMD Radeons. So… no chance on getting by with a Core i7?

I could rent out on AWS but it seems harder to learn when you have to do everything over the wire.

So… should I think about building a PC? A GTX 1050 seems to be the cheapest GPU with 4GB of VRAM. (Of course, not sure which motherboard/CPU and how much RAM is necessary yet).

If you can afford it, your own PC will be best on the long run. Do a search on the forums here, I remember seeing some very good guides for building one.

I will do a search, I already skimmed a thread. Problem is, most of these threads assume US prices and availability. It’s a whole other problem which of this stuff is best for Romania.

More than how powerful, it is to have a GPU with enough memory. I use for face recognition with convolutional networks not less than 6GB. With 4GB I can run only smaller models. Wit 2GB you cannot really work.
Also, if you like Linux, try to verify if the configuration you choose is compatible with the new 18.04 LTS.

1 Like

Hm, a GTX 1060 with 6GB VRAM is almost twice the price of the 4GB variant on pcgarage.

Clearly @3DSF_LTD has many external GPUs while @Virgil has a desktop. What is everybody else using?

I’ll actually try to do the main coursework using a free cloud service (either kaggle kernels or google colab).

1 Like

Hi! We’ll have the first fastai study group on Friday the 26th of October at 7PM at Cowork (Casa Tineretului Timisoara). We’ll have the place for ourselves for the night so we can take our time to discuss, set-up frameworks & code as much as we want.

3 Likes

Found the first lesson in fast.ai github repo (work in progress I think) and managed to run it in Google Colab with GPU enabled, after minor changes.
gist here : https://colab.research.google.com/gist/virgilpetcu/c60a7122fb9b58aaf790681254087989/lesson1-pets-working-in-g-colab.ipynb

So it looks like colab is a viable option

3 Likes

Hi,

I’ve managed to get to the live YouTube video, but have no access to the part3 fast.ai section.
Didn’t see the any “acceptance” email apart from that accessing the fast.ai forum.

Is this normal ? Or does this mean that I’m not officially picked for this live course session ?

Bogdan

If you have no access to part3 forum and didn’t receive the email containing the youtube link then yes, you probably were not officially selected for the live course.

I’ve searched the forum regarding Google Cloud Platform’s 300$ free credit tier and it seems that, since 1st of April, GPU is not included anymore:

As far as you can see in the upper part of the thread, prior 1st of April (no joke) GPU could be simply added to the free tier.