Manage generators
You need to be the owner or have Editor access to a generator to modify it.
Status of generators
The status of a generator indicates its current state.
Status | Description | Next actions |
---|---|---|
A generator object exists with a default or modified configuration. Training not started. | • Start training • Clone configuration • Delete | |
Generator training is queued until cluster resources become available. | • Share • Clone configuration • Cancel training • Delete | |
Generator training is ongoing. | • Share • Clone configuration • Cancel training • Delete | |
A trained generator was cloned to improve its quality with further training. Training not started. | • Start training • Share • Clone configuration • Delete | |
The generator has completed training successfully and can now generate synthetic datasets. | • Generate data • Share • Export to file • Clone configuration • Continue training • Delete | |
The generator training started and then failed. | • Share • Clone configuration • Delete | |
The generator training was canceled while still in progress. | • Share • Clone configuration • Delete |
Clone a generator
If you need to reuse the data and model configuration from an existing generator, you can clone it. All previously added data as well as the model and training configuration are copied to the new generator.
Before you start, keep in mind:
- Cloning is available only for generators that use a database or a cloud storage connector as a data source.
- You cannot clone generators with uploaded files because the uploaded data is deleted after the generator training completes.
Clone a generator from the Web UI, by following the steps below.
Steps
- Clone a generator directly from the Generators page.
- From the Generators page, click the kebab menu of a generator, and select Clone configuration.
- Clone a generator after you open it.
- From the Generators page, click a generator to open it.
- Open the actions menu by clicking the generator name.
- Click Clone configuration.
Result
A new generator with the name of the original generator with the prefix Clone - prepended to it is created.
What’s next
You can now use the data and model configuration from the previous generator and make any necessary changes before starting its training.
Continue training
There may be cases where you need to improve the quality of a generator by resuming its training from the current weights of the model. An example is improving the overall accuracy of the generator. In such cases, you can use the Continue training option.
Prerequisites
- The generator you want to improve must have already completed training successfully. You cannot improve generators with Failed or Canceled status.
- You can only improve generators that use a database or a cloud storage bucket as a data source. Uploaded files used for generator training are deleted immediately after training and therefore cannot be used to continue generator training.
- The source data must be available in the data source.
- You must have the Editor role.
You can continue generator training from the Platform or with the Synthetic Data SDK.
To continue generator training, follow these steps.
Steps
- Continue the training of a generator in one of two ways.
- From the Generators page, click the kebab menu of a generator, and select Continue training.
- From the generator page, open the action menu by clicking the generator name, and select Continue training.
The generator is cloned with the status
CONTINUE
. You can now configure the model and training options. The generator name is prefixed with Continue training - followed by the name of the original generator.
- (Optional) Click a table to expand its model and training configuration to adjust as needed.
- Click Continue training to start the generator training.
Result
MOSTLY AI fetches the original data from the data source and continues the training from the already saved model weights. The generator status is updated to CONTINUE
.
What’s next
You can use the newly trained generator to generate a new synthetic dataset or probe it for immediately generated samples.
Share a generator
You can share a generator with members of your organization, see Manage Resources.
Delete a generator
A generator consists of a dedicated generative AI model for each table of the source dataset. Depending on the size of your original data, it can take a long time to train a new one.
If you need to delete a generator, you can do so after you open a generator.
- Delete a generator in one of two ways.
- From the Generators page, click the kebab menu of a generator, and select Delete.
- From the generator page, open the action menu by clicking the generator name, and select Delete.
Result
The generator is now deleted.