diff --git a/aml/deployment.yaml b/aml/deployment.yaml index 00a2b15e..102292e1 100644 --- a/aml/deployment.yaml +++ b/aml/deployment.yaml @@ -1,6 +1,6 @@ $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json name: bloom-deployment -endpoint_name: bloom-azure-inference +endpoint_name: bloom-inference model: azureml:bloom-safetensors:1 model_mount_path: /var/azureml-model environment_variables: @@ -24,13 +24,13 @@ request_settings: request_timeout_ms: 60000 max_concurrent_requests_per_instance: 256 liveness_probe: - initial_delay: 80 + initial_delay: 140 timeout: 60 period: 60 success_threshold: 1 failure_threshold: 2 readiness_probe: - initial_delay: 80 + initial_delay: 140 timeout: 60 period: 60 success_threshold: 1 diff --git a/aml/endpoint.yaml b/aml/endpoint.yaml index ed769dcd..f2f01d5e 100644 --- a/aml/endpoint.yaml +++ b/aml/endpoint.yaml @@ -1,3 +1,3 @@ $schema: https://azuremlsdk2.blob.core.windows.net/latest/managedOnlineEndpoint.schema.json -name: bloom-azure-inference +name: bloom-inference auth_mode: key