docker/docker-compose.yml

# Example docker-compose for running the DSV4-Flash MTP vLLM image locally.
# Edit MODEL_HOST_PATH to point at the HF snapshot you've downloaded.
version: "3.9"

services:
  dsv4-flash:
    image: dsv4-flash-acti-mtp:0.1.0
    container_name: dsv4-flash
    restart: unless-stopped
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              capabilities: ["gpu"]
              count: all
    shm_size: "16gb"
    ulimits:
      memlock: -1
      stack: 67108864
    ports:
      - "8000:8000"
    volumes:
      - ${MODEL_HOST_PATH:-/home/ripper/dsv4-local/models/DeepSeek-V4-Flash-Acti-MTP-W4A16-FP8-GPTQ}:/models/DeepSeek-V4-Flash-Acti-MTP-W4A16-FP8:ro
      - dsv4_cache:/root/.cache
    environment:
      MODEL_PATH: /models/DeepSeek-V4-Flash-Acti-MTP-W4A16-FP8
      # Override any of the Dockerfile defaults here, e.g.:
      # MAX_MODEL_LEN: 131072
      # MAX_NUM_SEQS: 8
      # GPU_MEMORY_UTILIZATION: 0.90

volumes:
  dsv4_cache: