FAQ about the course projects

[lewtun]

Hello everyone !

I’ll collect here any frequently asked questions that arise during the course of the community event on November 15th - 19th. Feel free to comment here if something is unclear

Can I propose my own project?

Yes, this is certainly encouraged! Just create a new topic under the #course:course-event category. You can use one of the existing project ideas as a template for what information to include.

Can I work on more than one project?

Yes, but we recommend completing one first with before starting/joining another.

How do I join a team?

Just comment on one of the project ideas in #course:course-event to indicate that you’d like to contribute. Alternatively, you can propose your own project idea by creating a new topic on the forums. Just tag me (@lewtun), so I know what you’re proposing

How many people can each team have?

Each team can have 1-4 people. We’ve found in previous events that having more than 4 people adds planning complexity. If you want to work by yourself, please note that you’ll have to use your own GPUs / cloud provider as the Amazon SageMaker compute is reserved for teams with 2 or more people.

What should I do after joining a team?

There are three main things you should do after joining a team:

1. Head over to Discord and join the project’s channel. If a channel does not yet exist, you can request it in the #request-a-channel channel and someone in the team will create it for you.
2. Join the Amazon Sagemaker Community organisation on the Hub. Every model trained using the free compute on Amazon SageMaker will need to be uploaded to that organization.
3. [Optional] Join your team’s organisation on the Hub. The organisation can be used to store your datasets and Streamlit / Gradio demo as a Space. If an organisation doesn’t yet exist for your project, just create one yourself and share the link in the project topic.

What are the requirements to obtain the course certificate?

Your team should have a Streamlit or Gradio demo running on Spaces. There should only be one Space per team, and you can deploy it under a single user account or organization.

Someone from the team will check that the application runs and then everyone in the team will receive their certificate. The deadline is November 24th.

Where should our team deploy our Space?

You can deploy your team’s space in several locations:

• Under a single team members username (e.g. here’s a Space created under my account)
• Under an organisation that you create for the team (e.g. here’s a Space created under the course org)
• Under the Amazon SageMaker Community org (here’s an example)

If you want to transfer a Space from your user account to an organisation, just select the Settings tab of your Space and scroll down to the Rename or transfer this space box:

There you can select New owner to change the location of your Space to an organisation.

My team has deployed a Space for our project, what do we do next?

1. Make sure all the team members (i.e. Hub usernames) are listed on the Space
2. Post a link to your Space in the corresponding project topic in #course:course-event or project channel in Discord. Feel free to tag @lewtun so we don’t miss it
3. [Optional] Write a nice model card for the models that you trained for the Space
4. Celebrate on completing the project !

How can I access the compute provided by Amazon SageMaker?

1. Fill out this registration form
2. Request to join the Amazon SageMaker Community organisation on the Hub

Do I have to use Amazon Sagemaker to train my models?

Although we encourage you to use the free compute generously provided by Amazon Sagemaker, you are welcome to use your own GPUs / cloud provider.

What kind of compute resources will be supplied by Amazon Sagemaker?

You can expect to have access to a single T4 GPU (16 GB RAM) between November 17-19. After November 19 you’ll need to use your own compute. Fortunately, there are several free options you can try:

• Google Colab
• Kaggle notebooks
• Paperspace Gradient notebooks

Can I share my Amazon SageMaker credentials with another team member?

No, please fill out the registration form and you’ll receive a confirmation email with instructions sometime later.

[MoritzLaurer]

What kind of compute resources will be supplied by AWS? Which type, how much RAM, for how long can we run it?

Background: I have some ideas for projects, but they require training a large model which cannot be trained with a normal single google colab pro T4/P100 GPU (for training e.g. a DeBERTa-large model even the ‘high-RAM’ GPUs from colab are not enough).

[sgugger]

I don’t think the GPUs will be more than T4s on AWS SageMaker either (and only one GPU not several), which is why we recommend projects that involve fine-tuning pretrained models more than pretraining.

[nouamanetazi]

Hello, can we participate in more than one project/team? (there are so many interesting projects )

[sgugger]

Complete one first, then you can join another if there is room

[Alifarsi]

Hi, so I created the space and my app is running. how can I add other team members so they can push their code and collaborate?

[sgugger]

You need to put your space in an organization where your team members also are. The Amazon SageMaker Community org, or an org specific to your project you have to create

[Alifarsi]

I created a non-profit org, how can I add my space to org?

[sgugger]

You should be able to move the space you created under your username to that org, in the settings.

[Alifarsi]

I couldn’t find it in space or org settings.

[sgugger]

In the screenshot that you pasted, scroll the menu under “New owner” and you will be able to transfer it to the organization you created.

[Alifarsi]

oh I see…thanks for saving the day Sylvain!