Sentence-Transformers Cheatsheet

Q: How do I choose a model?

all-MiniLM-L6-v2 (fast, good quality, 384-dim) is recommended for most use cases. all-mpnet-base-v2 (slower, better quality, 768-dim) is best for accuracy-critical tasks. multilingual models like paraphrase-multilingual-MiniLM-L12-v2 support 50+ languages.

Q: What's the difference between Bi-Encoder and Cross-Encoder?

Bi-Encoder encodes each text independently into a vector — fast, suitable for retrieval (thousands of candidates). Cross-Encoder processes text pairs together — slower but more accurate, suitable for re-ranking the top 10-100 results. Use Bi-Encoder for first-pass retrieval, Cross-Encoder for second-pass re-ranking.

Sentence-Transformers is a Python library for encoding text into dense vectors using pretrained Transformer models. Widely used for semantic search, clustering, text similarity, and RAG embeddings. Supports both Bi-Encoder and Cross-Encoder architectures.

Updated: 2026-07-20·7 commands

Quick Start

``bash pip install sentence-transformers

python -c " from sentence_transformers import SentenceTransformer model = SentenceTransformer('all-MiniLM-L6-v2') emb = model.encode('Hello world') print(emb.shape) # (384,) "``

Startup & Modes(1)

Command	Level	Example
`pip install sentence-transformers` Install Sentence-Transformers	Basic	`pip install sentence-transformers`

ai-embedding(6)

Command	Level	Example
`st encode` Encode text into embedding vectors	Basic	`python -c 'from sentence_transformers import SentenceTransformer; model = SentenceTransformer("all-MiniLM-L6-v2"); emb = model.encode("Hello world"); print(emb.shape)'`
`st encode batch` Batch encode multiple texts	Basic	`python -c 'from sentence_transformers import SentenceTransformer; model = SentenceTransformer("all-MiniLM-L6-v2"); embs = model.encode(["Hello", "World"]); print(embs.shape)'`
`st similarity` Compute semantic similarity between two texts	Basic	`python -c 'from sentence_transformers import SentenceTransformer, util; model = SentenceTransformer("all-MiniLM-L6-v2"); emb1 = model.encode("I love cats"); emb2 = model.encode("I like dogs"); print(util.cos_sim(emb1, emb2))'`
`st paraphrase mining` Find semantically similar sentence pairs	Expert	`python -c 'from sentence_transformers import SentenceTransformer, util; model = SentenceTransformer("all-MiniLM-L6-v2"); sentences=["I love cats","I like dogs","Cars are fast"]; embs = model.encode(sentences); pairs = util.paraphrase_mining(embs, sentences); print(pairs[0])'`
`st semantic search` Search most relevant texts in a corpus	Intermediate	`python -c 'from sentence_transformers import SentenceTransformer, util; model = SentenceTransformer("all-MiniLM-L6-v2"); corpus=["AI is great","I love programming"]; query="machine learning"; c_emb = model.encode(corpus); q_emb = model.encode(query); print(util.cos_sim(q_emb, c_emb))'`
`st cross-encoder` Use Cross-Encoder for re-ranking	Expert	`python -c 'from sentence_transformers import CrossEncoder; model = CrossEncoder("cross-encoder/ms-marco-MiniLM-L6-v2"); pairs=[("Query","Doc1"),("Query","Doc2")]; print(model.predict(pairs))'`

FAQ

How do I choose a model?

What's the difference between Bi-Encoder and Cross-Encoder?

This cheatsheet is compiled from official tool documentation. Last updated: 2026-07-20.