Revision Workflow Study Example

This example demonstrates how CARE can support a structured revision workflow to study how users revise and respond to AI-generated feedback. It enables natural editing of a given document, tracks changes, and summarizes them for further analysis.

The workflow follows a four-step process:

1. Initial Revision – users revise a pre-written text.

2. LLM Feedback – model-generated suggestions are shown.

3. Second Revision – users revise again, now with feedback.

4. Final Summary with LLM Feedback – feedback and changes are presented in a modal.

This workflow is available in the CARE interface under the name “Ruhr-Uni-Bochum-Project”.

Prepare external GPU Provider

• Use Leaf.Cloud setup guide

Prepare the broker

Install broker locally

cd /home/deployer
git clone https://git.ukp.informatik.tu-darmstadt.de/zyska/CARE_broker.git my_broker
git checkout <your-branch-name> # needs to be added on the server
cd my_broker
nano .env
make docker

.env:

PROJECT_NAME=my_broker
BROKER_PORT=4858

Prepare the CARE Instance

1. Ask Sysadmin Team to add A-Record for new domain (in this case rub.ukp.informatik.tu-darmstadt.de)

2. Clone branch into subfolder and install CARE instance

cd /home/deployer
git clone https://git.ukp.informatik.tu-darmstadt.de/zyska/care.git my_care
cd my_care
git checkout <project-branch>
git checkout -b deploy-<your-name>
nano .env
git add .env
git commit
git push --set-upstream origin deploy-<your-name>
make build

.env:

PROJECT_NAME=my_care
PUBLISHED_PORT=8992
POSTGRES_URL=postgres
RPC_MOODLE_HOST=rpc_moodle
RPC_MOODLE_PORT=8081
ADMIN_PWD=PjGcbLEUDEH93PY

3. Update NGINX and Certificates

Note

Make sure that the DNS Entry is available outside of the TuDa (may need some time)

cd /srv/nginx-certbot
sudo nano init-letsencrypt.sh # add domain here
sudo ./init-letsencrypt.sh # reinit

cd /srv/nginx-certbot
sudo nano data/nginx/app.conf
docker compose restart

data/nginx/app.conf:

# For your instance
server {
    listen 80;
    server_name rub.ukp.informatik.tu-darmstadt.de;
    server_tokens off;

    location ^~ /.well-known/acme-challenge/ {
        root /var/www/certbot;
    }

    location / {
        return 301 https://$server_name$request_uri;
    }
}

server {
    listen 443 ssl;
    server_name rub.ukp.informatik.tu-darmstadt.de;
    server_tokens off;

    ssl_certificate /etc/letsencrypt/live/care.ukp.informatik.tu-darmstadt.de/fullchain.pem;
    ssl_certificate_key /etc/letsencrypt/live/care.ukp.informatik.tu-darmstadt.de/privkey.pem;
    include /etc/letsencrypt/options-ssl-nginx.conf;
    ssl_dhparam /etc/letsencrypt/ssl-dhparams.pem;

    # if some service is not available
    error_page 501 502 503 /maintenance.html;
    location = /maintenance.html {
        root /var/www/html;
    }
}

For deployment and server configuration (e.g., NGINX, certificates), please refer to the general CARE installation guide: ../getting_started/installation.

Testing

1. Make sure the leaf.cloud server is running.

2. Connect to VPN

3. Run CARE locally with the latest branch

4. Update the settings (service → url) with your public broker URL

5. Restart backend

---

Leaf.Cloud

There are many external service providers for conducting studies with GPU resources, but only a few provide server capacities inside the EU allow accounting according to time of use. One of them is:

https://leaf.cloud/ (up to four Nvidia V100)

This tutorial provides only the basics for using a GPU with CARE. For additional functionality, see the documentation of Leaf.Cloud under https://docs.leaf.cloud/en/latest/. For price calculation, see https://www.leaf.cloud/pricing/.

Prepare Leaf.Cloud

Note

It is assumed that an account already exists and money has been transferred to it.

For all steps, login to the OpenStack Dashboard (https://create.leaf.cloud/) first.

Change language of the Dashboard

Click on Settings on the top right and change the Language to English (en). Click on Save to apply the change.

Create Key Pairs for Server Login

1. Click on Compute → Key Pairs → Create Key Pair

2. Fill out the form:

Field	Description	Example
Key Pair Name	Name of the key pairs	Login
Key Type	Type of the key pairs	SSH Key

3. After clicking Create Key Pair, a .pem file should download.

Note

Of course, you can also create the key pairs externally and upload the public key using Import Public Key.

Create Instance

Create Volume

Create a new disk under Volumes → Volumes → Create Volume

Field	Description	Example
Volume Name	Name of the disk	Study Data
Description	Description of the disk	Space for LLM Model and Study Data
Volume Source	Where to copy the disk from	No source, empty volume
Type	Type of the disk	encrypted
Size	Size of the disk	150 GiB
Availability Zone	Where the disk should be placed	europe-nl-ams1
Group		No Group

Create Instance

Create an instance under Compute → Instances → Launch Instance

Field	Description	Example
Instance Name	Name of the instance	Study Server
Volume Size (GB)	Size of the disk	20
Source Allocated	Operation System	Debian-12
Networks	Communication channel	external
Flavor	Server Capacity	eg1.a100x1.V12-84
Key Pair	Key Pair for SSH access	Server-Login

With Launch Instance the instance will be created. This takes around two minutes.

Important

Note down the IP address.

Attach Volume

Click on the previously created instance, Attach Volume, and select the volume with the study data.

Working with the Instance

You need the private key file (login.pem) and the IP address of the instance.

Login per SSH

chmod 0600 login.pem
ssh -i login.pem debian@<ip>

Mount Volume

sudo apt update
sudo apt install fdisk
sudo fdisk -l
# sudo fdisk /dev/vdb
sudo mkfs.ext4 /dev/vdb  # only once when the volume is created
sudo mount /dev/vdb /srv

# Mount on boot
sudo blkid  # copy UUID from /dev/vdb
sudo nano /etc/fstab
# Add: UUID=<copied UUID> /srv ext4 defaults 0 0

Install NVIDIA Driver

Add to /etc/apt/sources.list:

deb http://deb.debian.org/debian/ bookworm main contrib non-free non-free-firmware

and execute:

sudo apt update
sudo apt install nvidia-driver firmware-misc-nonfree
sudo reboot
nvidia-smi  # check if GPU is available

This should look like this:

Fri Feb  7 08:43:10 2025
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.216.01             Driver Version: 535.216.01   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA A100 80GB PCIe          On  | 00000000:00:05.0 Off |                    0 |
| N/A   48C    P0              49W / 300W |      0MiB / 81920MiB |      0%      Default |
|                                         |                      |             Disabled |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+

Install Conda

cd /srv
sudo curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
sudo chmod +x Miniconda3-latest-Linux-x86_64.sh
sudo bash Miniconda3-latest-Linux-x86_64.sh
# change location to /srv/miniconda3
sudo su -
conda info

Install broker for testing

Note

For security, configure the firewall to only allow inbound requests from specific IP addresses, especially before starting any network services like the broker.

Install Docker

From https://docs.docker.com/engine/install/debian/#install-using-the-repository :

sudo apt-get update
sudo apt-get install ca-certificates curl
sudo install -m 0755 -d /etc/apt/keyrings
sudo curl -fsSL https://download.docker.com/linux/debian/gpg -o /etc/apt/keyrings/docker.asc
sudo chmod a+r /etc/apt/keyrings/docker.asc

echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] \
https://download.docker.com/linux/debian \
$(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
sudo tee /etc/apt/sources.list.d/docker.list > /dev/null

sudo apt-get update
sudo apt-get install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
sudo docker run hello-world

Install BrokerIO

sudo apt install git
git clone https://github.com/UKPLab/CARE_broker.git /srv/brokerio
cd /srv/brokerio
docker compose -f docker-compose.yml -p "brokerio" up --build -d
docker ps

The broker should now be running on 127.0.0.1:4852.

IMPORTANT! Shelve Instance

Important

To save money, after usage the server shut be shelved.

1. Open the OpenStack Dashboard (https://create.leaf.cloud/)

2. Click on Compute → Instances and select the Study Server

3. On the right is the Actions Panel, click the small arrow

4. Click on Shelve Instance

The instance status should be Shelved Offloaded now! Make sure this is the status.

If you want to use the instance again, click in the same Actions panel on Unshelve Instance.