Add an NLU model to an update set
Use update sets to move your Natural Language Understanding (NLU) models from one instance to another. Update sets include all records needed for your model to function on the target instance.
始める前に
- Make sure that the NLU Workbench plugin, NLU Workbench - Core plugin, and Predictive Intelligence plugin are all installed and activated on your instance.
- The following instructions apply to NLU models for Virtual Agent and AI Search.
- Role required: admin
このタスクについて
Use update sets to transfer a model from a source instance to a target instance. The target instance must already have the same scope as the model's scope in the source instance.
- Model content (intents, utterances, entities, annotations, vocabulary, and the default test set)
- Associated vocabulary sources
- Corresponding latest active ML solution
- ML model artifacts
- ML solution and definition (last three runs, provided one of them was successful)
For more information on update sets, see System update sets.
If you want to add the model to a different update set, you must make that update set the current one. See Create and select an update set as the current set.
For optimum portability, add your NLU model to a new, dedicated update set rather than to a system Default update set. To do this, a new update set must be manually created in the model's scope. An error is displayed if the current update set is the scope's Default. This error contains a link for manually creating a new update set.
If the model contains records from multiple scopes, such as vocabulary tables, its update set must have a parent-child (batch) structure. Follow the procedure from step 5 to create a parent-child update set.
When models (either Global or scoped) are moved using update sets, their training and publishing state remains the same after transfer. So a model that is trained and published before it is added to an update set does not require retraining or republishing on the target instance.
For information about parent-child update sets, see Working with batched update sets. The following video provides a demonstration of this process.
手順
次のタスク
If a referenced record in the source instance is not present on the target instance, you may encounter errors while applying the update set. For example, the origin field on sys_nlu_intent could refer to an intent from a different model that doesn't exist on the target instance. You can click Accept remote update on the failed records to commit the update set anyway.