hf_text-generation-inference/aml/deployment.yaml

$schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
name: bloom-deployment
endpoint_name: bloom-inference
model: azureml:bloom-safetensors:1
model_mount_path: /var/azureml-model
environment_variables:
  WEIGHTS_CACHE_OVERRIDE: /var/azureml-model/bloom-safetensors
  MODEL_ID: bigscience/bloom
  NUM_SHARD: 8
environment:
  image: db4c2190dd824d1f950f5d1555fbadf0.azurecr.io/text-generation-inference:0.2.0
  inference_config:
    liveness_route:
      port: 80
      path: /health
    readiness_route:
      port: 80
      path: /health
    scoring_route:
      port: 80
      path: /generate
instance_type: Standard_ND96amsr_A100_v4
request_settings:
  request_timeout_ms: 90000
  max_concurrent_requests_per_instance: 256
liveness_probe:
  initial_delay: 600
  timeout: 90
  period: 120
  success_threshold: 1
  failure_threshold: 5
readiness_probe:
  initial_delay: 600
  timeout: 90
  period: 120
  success_threshold: 1
  failure_threshold: 5
instance_count: 1