Setting Up MLMs
Before setting up MLM, you should install a container runtime capable of running the Open Container Initiative (OCI) container images. Docker is one such container runtime. The installation process depends on the operating system and OCI runtime.
We have tested and validated containerized services (including UDFs, Spark, TensorFlow, and ONNX) using Docker on Linux, and Docker Desktop and Podman on Windows.
Note: In client-server deployments, the model path must be accessible from the server, with read access to the model location and write access to required cache or temporary directories.
The process for setting up MLMs is as follows:
1. Download the container image as per the model type (TensorFlow/ONNX) and configure the server to use the containerized service either:
• During installation, by selecting the ‘-tflowdownload’ flag for TensorFlow and ‘-onnxdownload’ flag for ONNX.
OR
• After installation, by using the ‘iisutensorflow’ script for TensorFlow and ‘iisuonnx’ script for ONNX:
iisutensorflow -download_newest_compatible
iisuonnx -download_newest_compatible
2. Containerized services use a cache directory to cache models and mount them into containers. The cache location can be configured using the “udfcache” parameter in the server group in vectorwise.conf.
On Linux:
DEFAULT: /tmp/vw_udfcache
On Windows:
DEFAULT: %PROGRAMDATA%\Actian\vw_udfcache
Last modified date: 09/11/2026