Machine Learning: Issues Accessing ADI-Integrated Models on Azure Databricks — Endpoint Not Found

Hi @edith-2570
Thankyou for reaching microsoft Q&A! and for sharing the troubleshooting details. We understand that you’re trying to invoke the GLM 5.3 model in your Azure Databricks environment, and we appreciate you confirming that other non-ADI foundation models are working as expected.Based on our investigation, we would like to highlight a few points that may help identify the cause of the issue.

Although GLM 5.3 is visible in Unity Catalog with a READY status, this does not necessarily mean that the model is available for invocation from your workspace.

GLM 5.3 is backed by ADI Services, so its availability depends on whether ADI Services is enabled for your Databricks account and whether the model is supported in your workspace's region.

We also understand that your workspace is located in North Europe and that the Integration with ADI Services option is not visible in your environment. Additionally, the current regional availability information does not list GLM 5.3 for North Europe, although it is listed for certain other regions, such as West Europe.

To help narrow down the issue, we recommend checking the following with your Databricks account team:

1. Confirm whether your Databricks account is eligible for and enabled to use ADI Services.
2. Verify whether the Integration with ADI Services option should be available for your account.
3. Confirm whether GLM 5.3 is supported in the North Europe region.
4. Check whether the model is available for invocation from your specific workspace.
5. If the model is not supported in North Europe, explore whether deploying the workload in a supported region is an option.

For additional information, please refer to the Databricks documentation on ADI Services.

Thank you for your cooperation. We appreciate your time and patience while working through this issue.

Hi edith-2570 & thx for join me at Q&A platform,

missing ADI Services toggle is probably the biggest clue here. GLM 5.3 is available through ADI Services, which requires separate eligibility and enablement. Having a Premium workspace and Unity Catalog enabled doesn't automatically grant access to those models.

Also, READY in Unity Catalog doesn't necessarily mean the corresponding serving endpoint is available in ur workspace. For GLM 5.3, the documented pay-per-token endpoint is databricks-glm-5-3. I'd check whether that exact endpoint is accessible through the Model Serving UI or the Serving Endpoints API. If it returns ENDPOINT_NOT_FOUND, while regular Databricks foundation models work, that points more toward missing ADI access than a general inference problem.

The setting should be under the Databricks account settings accessible from the workspace. Make sure ur checking with an account admin, not just a workspace admin. Those roles aren't interchangeable, and being able to manage an Azure resource through ARM doesn't automatically give u Databricks account-level permissions.

More importantly, North Europe doesn't currently list databricks-glm-5-3 among its supported pay-per-token models in the published regional availability table. So even with ADI enabled, u shouldn't assume GLM 5.3 will work in that region. That's a separate limitation from account eligibility. <Cite refs={["turn993629search0","turn965080search0"]}/>

I'd contact the Azure Databricks account team or support and ask them to confirm both ADI Services eligibility and GLM 5.3 availability for ur North Europe workspace. Include the workspace ID, exact endpoint name, and request ID from the failed invocation. If North Europe isn't supported for this model, they'll need to confirm whether there's a supported regional alternative or an upcoming availability change.

I wouldn't recreate the workspace just yet. Neither the ARM deployment method nor Unity Catalog's READY status proves that this particular model endpoint should be available there

rgds,

Alex

p.s. if my answer is hepl - pls mark it as an answer