docker-compose.yml

# kimodo.cpp text-to-motion demo, packaged for Portainer.
#
# Generic deployment (see docs/docker_portainer.md):
#
#   Docker Compose, from a repository checkout:
#     docker compose up --build          -> http://localhost:8094
#
#   Portainer: build the image on the host
#     (docker build -t kimodo:latest .)  -> deploy a Web editor stack with
#     this file minus the build: key, or a Repository stack as-is.
#
# The CPU backend is the default because it works everywhere; with no
# usable Vulkan device in the container, GGML's Vulkan enumeration can
# abort instead of falling back.  See the GPU section of the guide for
# the two-line change that enables GPU inference.

services:
  kimodo:
    image: kimodo:latest
    build: .
    container_name: kimodo
    restart: unless-stopped
    ports:
      - "8094:8094"
    environment:
      # Space-separated motion models to download on first start:
      # soma-rp-v1.1 soma-seed-v1.1 g1-rp-v1 g1-seed-v1
      KIMODO_MODELS: "soma-rp-v1.1"
      # Force the CPU backend: works on any host.  Remove when enabling
      # GPU inference below.
      KIMODO_BACKEND: "cpu"
      # Text encoder variant fetched on first start: bf16, q8_0 (default),
      # q6_k, q5_k, q4_k or q4_k_m.  Smaller variants trade prompt
      # fidelity for download size and memory.
      # KIMODO_TEXT_QUANTIZATION: "q8_0"
      # Text-encoder memory knobs (defaults mirror upstream: 32-layer
      # graphs, 10 GiB residency ceiling).  On a 4 GB GPU, bound the
      # residency instead of disabling GPU passthrough, e.g.:
      # KIMODO_TEXT_RESIDENT_LIMIT_MIB: "512"
      # KIMODO_TEXT_LAYER_CHUNK: "8"
      # CPU threads for CPU inference (defaults to all visible cores).
      # KIMODO_THREADS: "8"
      # Optional: Hugging Face token, raises rate limits for the
      # first-start weight download on shared connections.
      # HF_TOKEN: "hf_..."
    # For GPU inference (needs nvidia-container-toolkit on the host and
    # a generated CDI spec, see the guide), uncomment both:
    # devices:
    #   - nvidia.com/gpu=all
    # NVIDIA_DRIVER_CAPABILITIES: "graphics,compute,utility"
    volumes:
      - kimodo-data:/data

volumes:
  kimodo-data: