sentence-bert-base-architecture-bert

IN premise — summaries/2026/08/24/reimers-2019-sentence-bert-sR-references.md

Created 2026-08-25T02:58:30+00:00

Sentence-BERT fine-tunes the BERT architecture (Devlin et al., 2018) with a siamese/triplet architecture and custom training objectives (triplet loss, cosine-similarity regression) on NLI data.

Summary

Sentence-BERT is BERT retrained with a two-tower network design and geometric loss functions on natural language inference data, which reshapes a general-purpose token-level model into one that outputs a single, comparable vector for each sentence. This is the key step that makes sentence-level semantic search, clustering, and similarity scoring practical rather than relying on BERT's raw token outputs.