hppRC/defsent

Leverages dictionary definition sentences as training signals to improve semantic understanding in BERT and RoBERTa models, enabling both sentence encoding and salient word prediction from input text. Provides multiple pooling strategies (CLS, mean, max) across base and large model variants, with pre-trained checkpoints available on Hugging Face. Integrates directly with PyTorch via a simple `encode()` API and benchmarks competitively on semantic similarity tasks (STS, SICK-R).

No commits in the last 6 months. Available on PyPI.