port: 8004
env:
  VLLM_MODEL: kristaller486/dots.ocr-1.5
  VLLM_MAX_TOKENS: 24000
  VLLM_TEMPERATURE: 0.2
  VLLM_TOP_P: 0.9
  VLLM_REQUEST_TIMEOUT: 3000.0
  VLLM_MAX_MODEL_LEN: 32768
  VLLM_WORKER_MULTIPROC_METHOD: spawn
  VLLM_LOGGING_LEVEL: INFO
  HF_HUB_ENABLE_HF_TRANSFER: 1
  HF_HOME: /root/.cache/huggingface
  VLLM_DOWNLOAD_DIR: /root/.cache/huggingface
  CUDA_LAUNCH_BLOCKING: 0
  LOG_LEVEL: INFO
  VLLM_PORT: 8005
  TRANSFORMERS_OFFLINE: 1

gpu:
  enabled: true
  nodeSelector:
    node-role.deckhouse.io/cuda-worker: ""
  tolerations:
  - key: dedicated.deckhouse.io
    operator: Equal
    value: cuda-worker
    effect: NoExecute

resources:
  _default:
    requests:
      cpu: "1"
      memory: "10Gi"
    limits:
      memory: "10Gi"
      nvidia.com/gpu: 1

storage:
  size:
    _default: 10Gi
  class:
    _default: local-storage-class-worker

ingress:
  className:
    _default: nginx
  host:
    _default: vllm.pro.lukoil.com
    preprod: vllm.preprod.pro.lukoil.com
    prod: vllm.prod.pro.lukoil.com
  secretName:
    _default: vllm-tls