BERTSentenceEncoder

Includes a custom implementation of BERT (Bidirectional Encoder Representations from Transformers), finetuned using the NLI (Natural Language Inference) dataset, constructing a classification-based model and a contrastive approach-based model. The three BERT models are evaluated using the STS (Semantic Textual Similarity) benchmark, and the Pearson & Spearman correlation scores are compared.

Open in colab.