Problem

 When Translation Quality Estimation (TQE) runs, segments are sent to the model in batches - i.e. several segments are bundled into a single request. The number of segments per request is currently
hardcoded to 10in the backend (QualityEstimateService::MAX_SEGMENTS_PER_REQUEST = 10) and there is no way for the user to change it. The same kind of setting
already exists for the AI (pre-translation) language resources, where the user can configure the batch size - how many segments are tried to be sent in one request - directly in the resource settings via the
Max. batch sizefield. TQE has no equivalent field. Because the value is fixed, the batch size cannot be tuned per resource/model - for example to fit a model's context window / token limits, to trade throughput against reliability, or to react to a model returning incomplete or faulty batch responses. Different TQE models need different batch sizes, and right now all of them are forced to 10.
 

Solution

 
Add a batch sizefield to the TQE language resource settings UI, analog to the existing
Max. batch size field of the AI resources, and use the configured value instead of the hardcoded constant. Concretely: - Add a numeric batch size field to the TQE language resource settings, mirroring the AI-resource field (maxBatchSize) in label, tooltip, validation and behaviour.