minimax 2.5 docker compose

docker-compose.yml

services:
  minimax:
    image: ghcr.io/ggml-org/llama.cpp:server-cuda
    container_name: minimax-m25
    restart: unless-stopped

    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility

    volumes:
      - ./models:/models

    ports:
      - "8080:8080"

    command: >
      -m /models/minimax-m2.5/minimax-m2.5-Q4_K_M.gguf
      --host 0.0.0.0
      --port 8080
      -ngl 999
      --ctx-size 24576
      --cpu-moe
      --no-warmup
      --parallel 1 
      --batch-size 1024 
      --ubatch-size 256
      --threads 32
      --threads-batch 32

  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    depends_on:
      - minimax
    restart: unless-stopped
    ports:
      - "3000:8080"
    environment:
      # WebUI expects OpenAI-compatible base
      - OPENAI_API_BASE_URL=http://minimax:8080/v1
      - OPENAI_API_KEY=dummy
      # optional but nice:
      - WEBUI_AUTH=false

to bootstrap:

curl -LsSf https://hf.co/cli/install.sh | bash
hf auth login
hf download ox-ox/MiniMax-M2.5-GGUF   --local-dir models/minimax-m2.5   --include "*.gguf"

run

docker compose up

optional (reduce disk swapping)

echo 'vm.swappiness=1' | sudo tee /etc/sysctl.d/99-llm.conf
sudo sysctl --system
sudo swapoff -a
sudo swapon -a