sbert-clustering-65-hours-vs-5-seconds

IN premise — summaries/2026/08/24/reimers-2019-sentence-bert-s6-ablation-study.md

Created 2026-08-25T02:58:29+00:00

Hierarchical clustering of 10,000 sentences takes ~65 hours with BERT cross-encoder (~50M pair computations) but ~5 seconds with SBERT pre-computed embeddings.

Summary

Clustering 10,000 sentences is essentially a two-day operation if you compute pairwise similarity on the fly with a cross-encoder, but collapses to a five-second job when you rely on pre-computed sentence embeddings. This roughly 46,000-fold gap is a hard engineering constraint: any pipeline that needs to cluster large text corpora must use the embedding route, or it becomes practically infeasible.