komodgn/meta-ml repository overview

⁠System Architecture: Model Loading Strategy

⁠1. Partial Pre-caching (Design)

The server implements a partial pre-caching strategy; core NLP assets and the emotion model are pre-loaded during the Docker build stage. Additional task-specific weights (e.g., RetinaFace for detection, gender analysis) are fetched on-demand to maintain an optimal balance between CI/CD build time and final image size.

⁠2. Hybrid Initialization (Behavior)

To ensure the initial deployment remains lightweight, the server utilizes a hybrid loading approach. While primary dependencies are pre-installed, certain high-precision ML models are initialized upon the first request. Please note that this may cause a one-time 'cold-start' delay during the very first analysis task.

⁠How to use this image

To pull and run the image:

$ docker pull komodgn/meta-ml:latest

$ docker run -d -p 8080:8080 \
  --env-file .env \
  -v "/path/to/your/google-key.json:/app/key.json" \
  --name meta-ml \
  komodgn/meta-ml:latest

Check the logs:

$ docker logs -f meta-ml

# Health Check Endpoint
GET http://localhost:8080/ping

⁠Environment Variables

Before running the container, create a .env file with the following configurations:

WEB_SERVER_URL=   
GOOGLE_VISION_API_KEY= 
GOOGLE_APPLICATION_CREDENTIALS=   
GOOGLE_CLOUD_QUOTA_PROJECT= 
PAPAGO_API_KEY_ID= 
PAPAGO_API_KEY= 

• WEB_SERVER_URL: URL of the Meta-Web server (e.g. http://host.docker.internal:8081⁠)
• GOOGLE_VISION_API_KEY: Your Google Cloud Vision API Key
• GOOGLE_APPLICATION_CREDENTIALS: Path to your Google Cloud Service Account JSON key (e.g. /app/key.json)
• GOOGLE_CLOUD_QUOTA_PROJECT: Your Google Cloud Project ID for quota management
• PAPAGO_API_KEY_ID: Naver Cloud Papago API Client ID
• PAPAGO_API_KEY: Naver Cloud Papago API Client Secret

Backend (FastAPI): backend⁠

Client (Android): application⁠