Quack96/aac-full-finetunes

AAC risk-aware fine-tunes Full fine-tunes produced for doctoral research on predictive text for augmentative and alternative communication (AAC). Each is a base model adapted on AAC-elicited text under one of two training objectives -- standard cross-entropy, or cross-entropy weighted by a per-utterance communicative-risk term -- across three seeds. These are research artefacts. They are small models fine-tuned on a modest corpus and are not suitable for deployment as… See the full description on the dataset page: https://huggingface.co/datasets/Quack96/aac-full-finetunes.