In v38, models for flows created in v36 and above are forward compatible, meaning that you can use them in v38 without having to retrain them during the upgrade process. As a result, forward-compatible models allow you to:
upgrade v36 or v37 models after upgrading the application to v38,
maintain the automation levels achieved in v36 or v37, and
continue existing training efforts for any use cases that may be in implementation when upgrading to v38.
Which models are forward-compatible
In v38, the following types of models are forward-compatible:
Locator models
Transcription models (i.e., sets of finetuning models)
Classification models
The following types of models are not forward compatible, so you need to train v38 versions of these models in order to use them in v38:
Text Classification models
To learn more about Text Classification, see Text Classification.
Forward-compatible models and upgrades
When upgrading to v38, keep in mind that some aspects of the v38-upgrade process differ from upgrades to previous versions.
How the upgrade process changes
If you're upgrading from v34 or earlier, you need to upgrade to v35 first and complete any submissions created in v34 or earlier before upgrading to v36, v37, or v38.
When upgrading to v38, you do not need to attach a v38 version of the trainer to your previous application version.
Models for flows created in v36 are not removed from the system during the upgrade process; they remain intact and work "out of the box" after upgrading to v37 or v38, with no retraining required.
In v38, Hyperscience supports flows and models created in v38, v37, or v36, but not those created in v35 or earlier.
If you want to use flows created in v35 or earlier, you need to upgrade those flows prior to upgrading to v38 — they are not supported in v38, nor are their models forward compatible.
Using v36 or v37 models and flows in v38
As mentioned earlier, you can continue using v36 and v37 flows in v38 without interruption or loss of automation. However, these flows can only use the models they were using before upgrading to v38, and you cannot import models for these flows. You cannot import a model from v35 or earlier for use with a v36, v37, or v38 flow.
If you need to change a model for a v36 or v37 flow (e.g., train it for a new Semi-structured layout), you must create a new model, along with a new flow in v38 to go with it. When changing a model, all of the models for the model's flow must be upgraded for the new v38 flow. In other words, you cannot upgrade a flow's models selectively or in a piecemeal fashion. You also cannot attach a v36 or v37 trainer to update the models for a v38 flow.
When using v36 or v37 flows, you can use cloned versions of the releases you used in v36 or v37. However, if you add Semi-structured layouts to those releases, there is no automation for the submissions matched to these layouts. That is, the system generates Supervision tasks for these submissions.
Memory Management
If you are using v36 of v37 flows and out-of-memory errors occur, you should enable the Memory Management feature. This feature assigns flows that were created in specific application versions to specific application machines in your instance.
To learn more about Memory Management, see Memory Management.
How to know when to train forward-compatible models
When you open a v36 or v37 flow in Flow Studio, the system lets you know which models for the flow need to be trained.
From there, you can click on an issue's Go to Model Management button.
If the affected model is a Locator Model, you are redirected to the Model Library. There, you can find the model mentioned in the issue's description and click on its name. On the model’s details page, you can take action to correct the issue.
If the model mentioned in the issue is a Classification or Transcription mode, clicking Go to Model Management takes you directly to the model’s details page, where you can take corrective action.