2023.2.1 Release Overview
We are pleased to announce the following product improvements in our 2023.2.1 release:
- Model Upload Expansion: With this release, Wallaroo expands the number of supported machine learning frameworks. ML Models that meet the supported model frameworks can be uploaded and deployed into Wallaroo workspaces and pipelines from an array of different model types and flavors.
- Supported frameworks include the following. See the reference documentation for full framework and version requirements.
- ONNX
- Python Shape Steps
- Tensorflow
- Keras
- Arbitrary Python
- Hugging Face
- PyTorch
- Sci-kit Learn aka SKLearn
- XGBoost
- Containerized MLFlow
- References:
- Wallaroo SDK Essentials Guide: Model Uploads and Registrations: Guides on uploading ML models to a Wallaroo instance using the Wallaroo SDK.
- Wallaroo MLOps API Essentials Guide: Model Upload and Registrations: Guides on uploading ML models to a Wallaroo instance using the Wallaroo MLOps API.
- Wallaroo ML Models Upload and Registrations Guides: Tutorials on uploading, registering, and deploying ML Models of various frameworks into a Wallaroo instance.
- Supported frameworks include the following. See the reference documentation for full framework and version requirements.
- GPU Pipeline Deployment Support: Pipeline configuration includes allocating GPUs to pipeline deployment configurations both native and containerized runtimes.
- MLFlow Registry Integration: Model Registry services connection details are made available to workspace users through the Wallaroo registry object. This provides methods of listing available models and their versions in a model registry, then uploading supported model artifacts into Wallaroo for pipeline deployments.
- Pipeline Log Reliability Enhancements: Wallaroo Pipeline Logs received the following updates.
- Python shape logging: Python scripts uploaded as ML Models that are constructed to accept pandas DataFrame or Apache Arrow inputs and outputs will have the same inference result and logging structures as other ML Models deployed in Wallaroo. This places the inputs and outputs under the
in
andout
metadata structures for consistent data input and retrieval. - Pipeline log reliability: Pipeline logs have a set storage space. To prevent log storage issues, inference result data may be restricted depending on the size of the input and output data. Inference results will always include all input and output details, while pipeline logs may drop extraneous data for storage purposes. The new update provides methods for developers to detect that information may be dropped from log storage and plan accordingly.
- References:
- Python shape logging: Python scripts uploaded as ML Models that are constructed to accept pandas DataFrame or Apache Arrow inputs and outputs will have the same inference result and logging structures as other ML Models deployed in Wallaroo. This places the inputs and outputs under the
- Arbitrary Python Model Support: Wallaroo ML Model support includes Arbitrary Python, providing organizations the ability to Python scripts with artifacts and libraries as ML models for deployment as Wallaroo pipeline steps. This provides a powerful level of flexibility in deploying models in a Wallaroo environment with additional support contained around them. Note that ML models included with an Arbitrary Python model must comply with Wallaroo model framework and versions requirements.