model_name: rime-mist-v3
model_metadata:
  description: >-
    Rime Mist v3 — low-latency conversational text-to-speech, served from Rime's
    Mist v3 Omni engine on a single H100. Closed-source: deploys
    from Rime's private engine image.
base_image:
  image: us-docker.pkg.dev/rime-labs/mist/v3/omni:20260620
  docker_auth:
    auth_method: GCP_SERVICE_ACCOUNT_JSON
    secret_name: gcp_rime_service_account
    registry: us-docker.pkg.dev
docker_server:
  no_build: true
  server_port: 8080
  predict_endpoint: /invocations
  readiness_endpoint: /readyz
  liveness_endpoint: /readyz
resources:
  accelerator: H100
  use_gpu: true
secrets:
  gcp_rime_service_account: null
  rime_license: null
  rime_api_key: null
environment_variables:
  RIME_LICENSE_FILE: /secrets/rime_license
  RIME_API_KEY_FILE: /secrets/rime_api_key
  MIST_TRT_CACHE_DIR: /cache/model/mist_v3_trt_cache
  MEASURE_GPU_LOAD_THREAD_ENABLE: "false"
runtime:
  health_checks:
    startup_threshold_seconds: 3000