version: "3.8"
services:
  ktransformers:
    image: docker.citory.tech/public/ktransformers:0.2.2rc1
    container_name: ktransformers
    runtime: nvidia
    tty: true
    stdin_open: true
    ports:
      - 10580:10580
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 2
              capabilities:
                - gpu
    ipc: host
    volumes:
      - /home/deepgeek/data/data_local/server/ktransformers/models:/workspace/models
      - /home/deepgeek/data/data_local/server/ktransformers/ktransformers/website:/workspace/ktransformers/ktransformers/website
    env_file:
      - .env
    restart: unless-stopped
    entrypoint: [
      "python3", "/workspace/ktransformers/ktransformers/server/main.py",
      "--gguf_path", "/workspace/models/DeepSeek-R1-GGUF/DeepSeek-R1-UD-Q2_K_XL",
      "--model_path", "/workspace/models/DeepSeek-R1",
      "--model_name", "deepseek-r1:671b",
      "--cpu_infer", "94",
      "--optimize_config_path", "/workspace/ktransformers/ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat-multi-gpu.yaml",
      "--max_new_tokens", "8192",
      "--cache_lens", "32768",
      "--total_context", "32768",
      "--cache_q4", "true",
      "--temperature", "0.6",
      "--top_p", "0.95",
      "--force_think",
      "--no-use_cuda_graph",
      "--host", "0.0.0.0",
      "--port", "10580"
    ]
x-dockge:
  urls:
    - http://local.citory.tech:10580
networks: {}