Update model serving endpoint

pablobd — Mon, 29 Jan 2024 12:48:50 GMT

Hi all,

I've been able to create a model serving endpoint through the api, following the docs, but then when trying to update the version, I get the following error:

'{"error_code":"RESOURCE_ALREADY_EXISTS","message":"Endpoint with name \'ml-project\' already exists."}'

Appreciate you help and insights.
Thanks

Re: Update model serving endpoint

Aaron12 — Tue, 30 Jan 2024 05:30:35 GMT

I was also searching for a solution to this problem when I came across your similar issue. This approach seems to work well. It either updates the config of the existing endpoint, or creates a new one if it doesn't exist.

# Define the endpoint config data = { "name": endpoint_name, "config": { "served_models": [ { "model_name": model_name, "model_version": model_version, "workload_size": "Small", "scale_to_zero_enabled": True, "environment_vars": { "OPENAI_API_KEY": f"{{{{secrets/{secret_scope_name}/{secret_key_name}}}}}" } } ] } } # Get list of active endpoints endpoint_list_response = requests.get( url=f"https://{db_host_url}/api/2.0/serving-endpoints", headers=headers ) # Check if endpoint already exists endpoints = endpoint_list_response.json().get("endpoints", []) endpoint_exists = any(ep['name'] == endpoint_name for ep in endpoints) print("Endpoint exists:", endpoint_exists) # Update endpoint config if it exists if endpoint_exists: update_response = requests.put( url=f"https://{db_host_url}/api/2.0/serving-endpoints/{endpoint_name}/config", json=data["config"], headers=headers ) print("Update Response status:", update_response.status_code) print("Update Response text:", update_response.text, "\n") # Create endpoint if it doesn't exist else: create_response = requests.post( url=f"https://{db_host_url}/api/2.0/serving-endpoints", json=data, headers=headers ) print("Create Response status:", create_response.status_code) print("Create Response text:", create_response.text, "\n")

Re: Update model serving endpoint

BR_DatabricksAI — Fri, 02 Feb 2024 08:57:30 GMT

Folks, Alternate way you can also deploy the models in serving layer with different versions. Though I am using mLflow.

You may also refer to the below link if it its helpful

How to Quickly Deploy, Test & Manage ML Models as REST Endpoints with Databricks - The Databricks Blog

Thanks.

topic Re: Update model serving endpoint in Machine Learning

Update model serving endpoint

Re: Update model serving endpoint

Re: Update model serving endpoint