HawkEyesAI/UBL-Backup

🐦‍🔥 Project: UNILEVER_Master → main Branch

HuggingFace Repo: https://huggingface.co/HawkEyesAI/UBL-Backup

🐳 Run with Docker (recommended)

Builds and runs on any machine with Docker Desktop — no GPU, no manual Python/apt setup. On a host with an NVIDIA GPU it also picks up CUDA automatically via the GPU override below.

One-time setup:

git lfs pull                      # pulls projectUNILEVER/AI_Models/*.pt (1.7GB)
cp .env.example .env              # optional — only needed to override defaults

Build and run:

# Any machine (CPU):
docker compose up -d --build

# Host with an NVIDIA GPU + NVIDIA Container Toolkit installed:
docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d --build

Confirm it's alive:

curl http://localhost:5656/status

Logs:

docker compose logs -f

(App-level rotating logs also land in ./logs/ on the host — bind-mounted, same files the app itself writes via projectUNILEVER/tools/logging_setup.py.)

Lint the Dockerfile/compose file and run a full build+health smoke test:

bash scripts/lint-docker.sh
bash scripts/docker-smoke-test.sh

Stop:

docker compose down

Publish to Docker Hub:

export IMAGE_NAME=yourdockerhubuser/ubl-api
export IMAGE_TAG=1.0.0
docker login
docker compose build
docker compose push

The image bakes in the full application source (COPY . . in the Dockerfile) — keep the Docker Hub repo private unless this code is meant to be public. AI model weights are never baked in (git-lfs, bind-mounted at runtime) — whoever pulls the image separately needs their own git lfs pull of projectUNILEVER/AI_Models/.

Single heavy inference process (10 YOLO models + facenet loaded into memory/GPU) — scale by running more containers behind a load balancer, not --scale/multiple uvicorn workers in one container, which would multiply memory for no throughput gain.

See Dockerfile, docker-compose.yml, and docker-compose.gpu.yml for details.

🖥️ Bare-metal Deployment (legacy)

The commands below run the app directly on the host (via tmux + ngrok) instead of in a container — kept for hosts that still use this flow.

🚀 Full Deployment Command

sudo apt update && sudo apt upgrade -y \
&& sudo apt-get install -y iproute2 libgl1 nano wget unzip nvtop git git-lfs \
&& git config --global credential.helper store \
&& git clone -b main https://huggingface.co/HawkEyesAI/UBL-Backup \
&& cd UNILEVER_Master \
&& chmod +x deploy.sh \
&& ./deploy.sh \
&& python test.py \
&& python countError.py

🧪 Manual Setup & API Run

git clone -b main https://huggingface.co/HawkEyesAI/UBL-Backup
cd UNILEVER_Master
pip install -r requirements.txt
sudo apt update && sudo apt upgrade -y
python UBL_API.py

🌐 Expose API via ngrok

tmux new-window -t deploy "ngrok http --domain=he.ngrok.app 5656"
tmux new-window -t deploy "ngrok http --domain=hawkeyes.ngrok.app 5656"

Direct command:

ngrok http --domain=he.ngrok.app 5656
ngrok http --domain=hawkeyes.ngrok.app 5656

⚡ Run with Hypercorn

tmux new-session -d -s deploy "python UBL_API.py"

or

tmux new-session -d -s deploy "hypercorn UBL_API:app --bind 127.0.0.1:5656 --workers 2"

📚 Citation

@misc{hawkeyes_digital_monitoring_ltd_2025,
    author       = { HawkEyes Digital Monitoring Ltd },
    title        = { UNILEVER_Master (Revision 6cbf0b8) },
    year         = 2025,
    url          = { https://huggingface.co/HawkEyesAI/UNILEVER_Master },
    doi          = { 10.57967/hf/7061 },
    publisher    = { Hugging Face }
}