all-MiniLM-L6-v2 is a public sentence-embedding checkpoint released through the sentence-transformers library. It is built on the nreimers/MiniLM-L6-H384-uncased backbone — a $6$-layer, hidden-size-$384$ Transformer produced by MiniLM-style self-attention distillation, with roughly $22.7$M parameters — and is fine-tuned with an SBERT-style contrastive objective on more than $1.17$ billion sentence pairs aggregated from Reddit comments, S2ORC citation pairs, WikiAnswers duplicates, PAQ, Stack Exchange, MS MARCO, and many additional sources. The model maps a sentence or short paragraph into a $384$-dimensional dense vector suitable for semantic search, clustering, and retrieval; inputs longer than $256$ word pieces are truncated. The released checkpoint is the canonical default encoder used in many downstream retrieval pipelines.