📢March 2023 second release
This release improved the model sharing and forking features, giving users more control over how you train your models. We also changed the default OCR engine to Azure.
Model forking & enhanced model sharing
We've introduced the ability to fork from model versions, which gives you more control over how you fine-tune your models. With this feature, you can choose your base model and train it further to meet your specific needs.
In addition, we've made some great improvements to our model sharing and forking features. Here's what you can expect:
You can now choose from which model version to start an incremental training
You can select the projects from which you want to use documents to launch a new training. If you don't have access to other projects, they will be listed under "Other projects".
You can now see on which documents a model version is trained, including some details about the projects of those documents (including F1 score of last trained model).
After training a new model, you need to explicitly share specific model version if you want other projects and organizations to use them.
Sharing of model versions cannot be undone if a project is using the model run.
With these changes, you have more control than ever before over how you train your models and work with shared models. We're confident that these new features will help you work more efficiently and effectively with our platform.
Default OCR changed to Azure
After extensive testing and experimentation we noticed that Azure comes out as best OCR engine when recognizing handwritten text, forms where characters are filled in boxes and general text recognition. That's why we have decided to change the default OCR engine to Azure. This change will not affect your current projects, only new projects will be impacted.
As always, we want to provide you with the flexibility to choose what works best for you. That's why you can always change your OCR engine at any time in the project settings.
Last updated
