We've just released new improvements!
Reduction in Autotuning Time
The time it takes for TVM acceleration workflows to complete is expected to decrease by up to 25% beginning this week.
Users do not need to do anything differently to access this improvement. This improvement will not change autotuning, benchmarking, packaging, or pricing.
ONNX Tarball format Upload Now Available
Users can now upload ONNX models in .tgz format, as long as the .tgz file includes a .onnx file within. Autotuning, benchmarking, and packaging work the same way on ONNX models that are uploaded as a .tgz file and those uploaded as a .onnx file.
In the Web UI, simply follow the existing model upload process: select the ONNX format, then either upload a file or drag and drop.
In the SDK, you will need to upgrade to the newest SDK version via
python3 -m pip install octomizer-sdk --extra-index-url [<https://octo.jfrog.io/artifactory/api/pypi/pypi-local/simple>](<https://octo.jfrog.io/artifactory/api/pypi/pypi-local/simple>) --upgrade
Customize the Name of your Packaged Model
You can now customize the module name of your package!
In the web UI, you will see a new optional field in the Accelerate UX.
To access the new model package naming in the SDK, you will need to upgrade to the newest SDK version via
python3 -m pip install octomizer-sdk --extra-index-url [<https://octo.jfrog.io/artifactory/api/pypi/pypi-local/simple>](<https://octo.jfrog.io/artifactory/api/pypi/pypi-local/simple>) --upgrade
The module name you use to call
importis now customizable to any Python standardized name you choose. Any irregular characters will be converted to an underscore. The system also now defaults to the model name you set when uploading a model, instead of a string likemodel_name_skylakein the past.The name of the Python wheel will continue to follow the same Python convention as in the past.
Email Notification
In the email notifications sent to users when a TVM acceleration workflow completes, users can now follow a link to the workflow in the web UI.