Skip to content

TRIAL.md

Latest commit

 

History

History
106 lines (78 loc) · 2.88 KB

TRIAL.md

File metadata and controls

106 lines (78 loc) · 2.88 KB
Raw

Trial Guide — Running AutoE2E Tests on AWS EC2

This guide walks you through launching a GPU instance, setting up the environment, and running the test suite end-to-end.

Step 1: Launch an EC2 Instance

AMI Selection

Use the Deep Learning OSS Nvidia Driver AMI (Ubuntu 24.04) provided by AWS.

To find it in the AWS Console:

  1. Go to EC2 → Launch Instance → Browse AMIs
  2. Search for: Deep Learning OSS Nvidia Driver AMI GPU PyTorch
  3. Select the latest Ubuntu 24.04 variant (e.g., Deep Learning OSS Nvidia Driver AMI GPU PyTorch 2.10 (Ubuntu 24.04))

Or via AWS CLI:

aws ec2 describe-images --owners amazon \
  --filters "Name=name,Values=Deep Learning OSS Nvidia Driver AMI GPU PyTorch 2.* (Ubuntu 24.04)*" \
            "Name=state,Values=available" \
  --query 'Images | sort_by(@, &CreationDate) | [-1].{Name:Name, ImageId:ImageId}' \
  --output table

Instance Type

Instance Type GPU vCPUs Memory Cost (us-east-1)
g4dn.xlarge (Recommended) 1× NVIDIA T4 (16 GB) 4 16 GB ~$0.526/h

Launch (CLI Example)

aws ec2 run-instances \
  --image-id <ami-id> \
  --instance-type g4dn.xlarge \
  --subnet-id <private-subnet-id> \
  --iam-instance-profile Name=<ssm-instance-profile-name> \
  --no-associate-public-ip-address \
  --block-device-mappings '[{"DeviceName":"/dev/sda1","Ebs":{"VolumeSize":100,"VolumeType":"gp3"}}]' \
  --tag-specifications 'ResourceType=instance,Tags=[{Key=Name,Value=auto-e2e-test}]'

Step 2: Connect via Session Manager

Once the instance is running and SSM agent is online (~30–60 seconds after launch):

aws ssm start-session --target <instance-id>

Then switch to the ubuntu user:

sudo -i -u ubuntu
cd ~

Step 3: Clone the Repository

git clone https://github.com/autowarefoundation/auto_e2e.git
cd auto_e2e

Step 4: Set Up Python Environment

The AMI comes with NVIDIA drivers pre-installed but Python packages are not available globally. Create a virtual environment:

python3 -m venv ~/e2e-env
source ~/e2e-env/bin/activate

Install dependencies with CUDA torch wheels (plain make setup installs CPU-only wheels):

make setup TORCH_CHANNEL=cu118

This takes 2–3 minutes. Verify the installation:

python -c "import torch; print(f'PyTorch {torch.__version__}, CUDA available: {torch.cuda.is_available()}')"

Expected output:

PyTorch 2.7.1+cu118, CUDA available: True

Step 5: Run the Tests

cd ~/auto_e2e
make test

The run ends with a pytest summary reporting that all selected tests passed. For the model's inputs, outputs and forward signature, see the Model architecture guide.

Step 6: Clean Up

Remember to terminate the instance when you are done to avoid unnecessary charges:

aws ec2 terminate-instances --instance-ids <instance-id>