SimCSE and RankCSE
SimCSE is a foundational contrastive learning approach for sentence embeddings that RankCSE builds upon and extends by incorporating ranking-based objectives to improve representation quality.
About SimCSE
princeton-nlp/SimCSE
[EMNLP 2021] SimCSE: Simple Contrastive Learning of Sentence Embeddings https://arxiv.org/abs/2104.08821
Provides both unsupervised and supervised training approaches—unsupervised leverages dropout-based noise on unlabeled data, while supervised incorporates NLI entailment pairs as positives and contradictions as hard negatives. Integrates seamlessly with HuggingFace Transformers and offers efficient similarity search via optional Faiss support, with pre-trained checkpoints available across BERT and RoBERTa architectures.
About RankCSE
perceptiveshawty/RankCSE
Implementation of "RankCSE: Unsupervised Sentence Representation Learning via Learning to Rank" (ACL 2023)
Scores updated daily from GitHub, PyPI, and npm data. How scores work