Train and Fine-Tune Sentence Transformers Models

Hugging Face Blog · 3 years ago

Sentence Transformers models were updated from the SentenceTransformer.fit API (pre-v3.0) to a new SentenceTransformerTrainer API with separate guides for embedding, reranker, sparse embedding, and multimodal models. The training process requires dataset preparation in one of four formats (sentence pairs with similarity labels, positive sentence pairs without labels, sentences with class labels, or triplet examples) and selection of an appropriate loss function matching the data structure. Users should now consult the updated guides covering the new SentenceTransformerTrainer rather than this outdated documentation.