# T5Gemma NetJets pronunciation-distribution adapter

QLoRA adapter for `google/t5gemma-2-4b-4b`, trained to predict NameCoach phonetic pronunciation distributions and deterministic mapped IPA.

This is an adapter, not a standalone base model. Load it with the matching T5Gemma base model and PEFT/Transformers. The adapter was initialized from the earlier grouped-mixture NetJets adapter and trained on a balanced occurrence/grouped mixture.

Evaluation artifacts and the exact data/split contract are in the companion dataset repository `Reza2kn/netjets-g2p-benchmark-artifacts`.