3.20.361.343 I slot print_timing: id  3 | task 465 | n_gen =   2037, tg =  23.75 t/s, tg_3s =  23.58 t/s
3.23.378.743 I slot print_timing: id  3 | task 465 | n_gen =   2108, tg =  23.74 t/s, tg_3s =  23.53 t/s
3.26.394.748 I slot print_timing: id  3 | task 465 | n_gen =   2179, tg =  23.74 t/s, tg_3s =  23.54 t/s
3.29.410.484 I slot print_timing: id  3 | task 465 | n_gen =   2250, tg =  23.73 t/s, tg_3s =  23.54 t/s
3.32.427.173 I slot print_timing: id  3 | task 465 | n_gen =   2321, tg =  23.72 t/s, tg_3s =  23.54 t/s


version: '3.8'

services:
  qwen-agent:
    image: ghcr.io/ggml-org/llama.cpp:server-rocm
    container_name: qwen-agent
    privileged: true
    ipc: host
    devices:
      - /dev/kfd
      - /dev/dri
    group_add:
      - video
    shm_size: '16gb'
    cap_add:
      - CAP_SYS_ADMIN
      - SYS_PTRACE
    security_opt:
      - seccomp=unconfined
    volumes:
      - /home/user/llm/containers/qwen/:/data
      - /home/user/llm/models/:/models
    ports:
      - "8001:8001"
    environment:
      - MODEL_PATH='/models/'
    command:
      - --model
      - /models/Qwen3.8-27B-UD-Q5_K_XL.gguf
      - "--port"
      - "8001"
      - "--host"
      - "0.0.0.0"
      - "--jinja"
      - "--chat-template-file"
      - "/data/chat_template.jinja"
      - "--ctx-size"
      - "228000"
      - "--flash-attn"
      - "on"
      - "--cache-type-k"
      - "q8_0"
      - "--cache-type-v"
      - "q8_0"
      - "--n-gpu-layers"
      - "999"
      - "--temp"
      - "1.0"
      - "--top-p"
      - "0.95"
      - "--top-k"
      - "20"
      - "--reasoning-budget"
      - "10240"
      - "--reasoning-budget-message"
      - "Proceed to final answer."
      - "--min-p"
      - "0.0"
      - "--repeat-penalty"
      - "1.0"
      - "--presence-penalty"
      - "1.5"