Parallel to OpenAI Assistants, we support OpenAI "Pseudo Training". Pseudo Training basically is a non-trainable Model, where a trainings system-message is sent alongside every translation request. To achieve this, the following adjustments have to be made (the basics are already there):

• A different Icon (Maybe a spanner without "Database") is added as the last action-icon behind non-trainable models.
• This icon is greyed out when no pseudo-training is attached, otherwise fully colored
• When the icon is clicked, a dialog opens to select the pseudo-training

Pseudo training window

• on top, the selected sysmessages are shown as list of their names with a button to remove them (with confirmation-dialog)
• Below is a button "Add system message" that opens a selector (the same as in the trainings window) to add a system-message

Data Model

There will be a new table to associate sysmessages with languageresources directly:

LEK_openai_pseudotraining_assoc

Holds the association between a languageresource and sys-messages. Generally, a training can have several sys-messages. The general presence of the training is captured in the specific data of the language-resource, see MittagQI\Translate5\Plugins\OpenAI\Api\Model\SPECIFICDATA_JOBKEY

When the prop "openAiPseudoTrained" in the specific data of the LR is set to 1, this indicates an active pseudotraining, otherwise 0. This steers also the icon being greyed out or not
   

   columns: ( id | languageresourceId | sysmsgId )