name: build-ml description: Build ML server Docker containers argument-hint: "[--push|--no-cache|model]"

ML Server Builder

Build Docker containers for ML server components.

Usage

• /build-ml - Build all ML containers
• /build-ml --no-cache - Build without cache
• /build-ml --push - Build and push to registry
• /build-ml finrl - Build specific model container
• /build-ml --cpu - Build CPU-only versions

ML Server Structure

services/ml-server/
├── docker-compose.yml      # ML services orchestration
├── docker-compose.dev.yml  # Development config
├── Dockerfile              # Base ML image
├── finrl/                  # FinRL deep learning
├── stockmixer/             # Stock mixing models
├── master/                 # Master orchestrator
├── samba/                  # Samba models
├── macrohft/               # Macro HFT models
└── src/                    # Shared ML code

Available Models

• finrl - Deep reinforcement learning (FinRL)
• stockmixer - Multi-asset mixing
• master - Model orchestration
• samba - Samba-based models
• macrohft - Macro HFT strategies

Instructions

When this skill is invoked:

1. Parse arguments:

   • --no-cache: Add --no-cache to docker build
   • --push: Push to container registry after build
   • --cpu: Use CPU-only base images
   • Model name: Build only that model's container

2. Navigate to ML server directory:

   cd services/ml-server
   

3. Build containers:

   # All containers
   docker compose build $NO_CACHE
   
   # Specific model
   docker compose build $MODEL $NO_CACHE
   

4. If --push specified:

   • Verify registry credentials
   • Tag images appropriately
   • Push to registry:

     docker compose push $MODEL
     

5. Report build results:

   • Build time per container
   • Final image sizes
   • Any build warnings
   • Layer cache efficiency

6. Common build issues:

   • CUDA not found: Use --cpu flag or install NVIDIA container toolkit
   • Out of memory: Increase Docker memory limit
   • Pip install fails: Check requirements.txt for version conflicts
   • Model weights missing: Download from model registry first

Image Tags

• latest - Most recent build
• v{version} - Release versions
• {git-sha} - Specific commit builds
• cpu - CPU-only builds