# Docker Compose x86_64 GPU Overlay
# Auto-applied by installer when NVIDIA GPU + Container Toolkit detected
# This overlay enables GPU passthrough for the vLLM embeddings container

services:
  vllm-embeddings:
    deploy:
      resources:
        # GPU mode: model loads into VRAM (~2.5GB), minimal system RAM needed
        limits:
          memory: 4G
        reservations:
          memory: 2G
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    environment:
      # GPU-specific settings
      - CUDA_VISIBLE_DEVICES=0
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    labels:
      - "gt2.gpu=enabled"
      - "gt2.gpu.vendor=nvidia"