all-MiniLM-L6-v2

all-MiniLM-L6-v2 is the most downloaded sentence embedding model on Hugging Face. It maps sentences and paragraphs into a dense 384-dimensional vector space, balancing speed and quality to be the default choice for semantic search, clustering, duplicate detection, and RAG retrieval.

Key features

• 384 dimensions — compact vectors that fit in memory and index fast
• High throughput — 6-layer transformer runs on CPU without GPU acceleration
• Strong performance — top-tier on STS and semantic search benchmarks for its size class
• Multilingual-adjacent — trained on diverse English data; pairs well with multilingual variants
• Drop-in ready — supported by sentence-transformers, LangChain, LlamaIndex, Chroma, FAISS, and Qdrant
• Tens of millions of downloads — the de facto default embedding model for OSS RAG pipelines

Quick start

npx ai-supply add all-minilm-l6-v2-embeddings

# Or install directly
pip install sentence-transformers

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")

sentences = [
    "The weather is lovely today.",
    "It's so sunny outside!",
    "He drove to the stadium."
]

embeddings = model.encode(sentences)
print(embeddings.shape)  # (3, 384)

# Compute similarity
from sentence_transformers.util import cos_sim
print(cos_sim(embeddings[0], embeddings[1]))  # High similarity
print(cos_sim(embeddings[0], embeddings[2]))  # Low similarity

Curated mirror of the open-source all-MiniLM-L6-v2 project (Apache-2.0). Install upstream from the repository.