Serving model issue in databricks

[Shivani_Pande]

I’m facing an issue while trying to deploy a custom pyfunc model for Qwen3-Embedding-8B (GGUF format) registered in Unity Catalog. The GGUF model file is stored inside a Unity Catalog Volume, and during model training and registration everything works correctly—logging completes successfully and a new version is created in the UC model registry. However, when I attempt to deploy the model via Databricks Model Serving, the deployment fails with the error: “Model path does not exist”, referring to the absolute path inside /Volumes/... where the GGUF file is located. The file is accessible and visible from an interactive notebook on a cluster (verified via listing and copying), but the serverless Model Serving environment appears unable to read from that location. The underlying issue seems to be that serverless serving does not have direct access to /Volumes paths referenced inside the model wrapper. I am seeking clarification on whether serverless model serving supports reading external files from UC Volumes, and whether the recommended approach is instead to log the GGUF file as a model artifact and load it via context.artifacts rather than referencing /Volumes directly.

[Advika · Accepted solution]

Hello @Shivani_Pande!

This issue appears to be very similar to another thread where accessing UC Volumes from Model Serving was discussed. In short, you need to use the Files API / SDK or package the files as model artifacts to read them from a Model Serving endpoint.
Check the detailed explanation here.

[iyashk-DB · Accepted solution]

Serverless Model Serving does not mount the UC Volumes FUSE path (/Volumes), so references to “/Volumes/…” inside a custom pyfunc’s model code will fail at container build or runtime. The correct pattern is to package any required files (like your GGUF) into the model artifact at log time and then load them from context.artifacts[...] in load_context() during serving.

Ref Doc - https://docs.databricks.com/aws/en/machine-learning/model-serving/model-serving-custom-artifacts

[Shivani_Pande]

Thank you so much. That solved my problem😀