sentence-transformers

Sentence Transformers - 最先进的嵌入

使用 transformers 进行句子和文本嵌入的 Python 框架。

何时使用 Sentence Transformers

在以下情况使用：

• 需要用于 RAG 的高质量嵌入
• 语义相似度和搜索
• 文本聚类和分类
• 多语言嵌入（100+ 种语言）
• 在本地运行嵌入（无需 API）
• 作为 OpenAI 嵌入的高性价比替代方案

指标：

• 15,700+ GitHub stars
• 5000+ 预训练模型
• 支持 100+ 种语言
• 基于 PyTorch/Transformers

改用其他替代方案的情况：

• OpenAI Embeddings：需要基于 API、最高质量
• Instructor：任务特定的指令
• Cohere Embed：托管服务

快速开始

安装

pip install sentence-transformers

基本用法

from sentence_transformers import SentenceTransformer

# Load model
model = SentenceTransformer('all-MiniLM-L6-v2')

# Generate embeddings
sentences = [
    "This is an example sentence",
    "Each sentence is converted to a vector"
]

embeddings = model.encode(sentences)
print(embeddings.shape)  # (2, 384)

# Cosine similarity
from sentence_transformers.util import cos_sim
similarity = cos_sim(embeddings[0], embeddings[1])
print(f"Similarity: {similarity.item():.4f}")

常用模型

通用

# Fast, good quality (384 dim)
model = SentenceTransformer('all-MiniLM-L6-v2')

# Better quality (768 dim)
model = SentenceTransformer('all-mpnet-base-v2')

# Best quality (1024 dim, slower)
model = SentenceTransformer('all-roberta-large-v1')

多语言

# 50+ languages
model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')

# 100+ languages
model = SentenceTransformer('paraphrase-multilingual-mpnet-base-v2')

特定领域

# Legal domain
model = SentenceTransformer('nlpaueb/legal-bert-base-uncased')

# Scientific papers
model = SentenceTransformer('allenai/specter')

# Code
model = SentenceTransformer('microsoft/codebert-base')

语义搜索

from sentence_transformers import SentenceTransformer, util

model = SentenceTransformer('all-MiniLM-L6-v2')

# Corpus
corpus = [
    "Python is a programming language",
    "Machine learning uses algorithms",
    "Neural networks are powerful"
]

# Encode corpus
corpus_embeddings = model.encode(corpus, convert_to_tensor=True)

# Query
query = "What is Python?"
query_embedding = model.encode(query, convert_to_tensor=True)

# Find most similar
hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=3)
print(hits)

相似度计算

# Cosine similarity
similarity = util.cos_sim(embedding1, embedding2)

# Dot product
similarity = util.dot_score(embedding1, embedding2)

# Pairwise cosine similarity
similarities = util.cos_sim(embeddings, embeddings)

批量编码

# Efficient batch processing
sentences = ["sentence 1", "sentence 2", ...] * 1000

embeddings = model.encode(
    sentences,
    batch_size=32,
    show_progress_bar=True,
    convert_to_tensor=False  # or True for PyTorch tensors
)

微调

from sentence_transformers import InputExample, losses
from torch.utils.data import DataLoader

# Training data
train_examples = [
    InputExample(texts=['sentence 1', 'sentence 2'], label=0.8),
    InputExample(texts=['sentence 3', 'sentence 4'], label=0.3),
]

train_dataloader = DataLoader(train_examples, batch_size=16)

# Loss function
train_loss = losses.CosineSimilarityLoss(model)

# Train
model.fit(
    train_objectives=[(train_dataloader, train_loss)],
    epochs=10,
    warmup_steps=100
)

# Save
model.save('my-finetuned-model')

LangChain 集成

from langchain_community.embeddings import HuggingFaceEmbeddings

embeddings = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-mpnet-base-v2"
)

# Use with vector stores
from langchain_chroma import Chroma

vectorstore = Chroma.from_documents(
    documents=docs,
    embedding=embeddings
)

LlamaIndex 集成

from llama_index.embeddings.huggingface import HuggingFaceEmbedding

embed_model = HuggingFaceEmbedding(
    model_name="sentence-transformers/all-mpnet-base-v2"
)

from llama_index.core import Settings
Settings.embed_model = embed_model

# Use in index
index = VectorStoreIndex.from_documents(documents)

模型选择指南

模型	维度	速度	质量	使用场景
all-MiniLM-L6-v2	384	快	好	通用、原型开发
all-mpnet-base-v2	768	中	更好	生产环境 RAG
all-roberta-large-v1	1024	慢	最佳	需要高精度
paraphrase-multilingual	768	中	好	多语言

最佳实践

1. 从 all-MiniLM-L6-v2 开始 - 良好的基线
2. 归一化嵌入 - 更适合余弦相似度
3. 如有 GPU 则使用 - 编码速度提升 10 倍
4. 批量编码 - 更高效
5. 缓存嵌入 - 重新计算成本高
6. 针对领域进行微调 - 提升质量
7. 测试不同模型 - 质量因任务而异
8. 监控内存 - 大型模型需要更多 RAM

性能

模型	速度（句/秒）	内存	维度
MiniLM	~2000	120MB	384
MPNet	~600	420MB	768
RoBERTa	~300	1.3GB	1024

资源

• GitHub：https://github.com/UKPLab/sentence-transformers ⭐ 15,700+
• 模型：https://huggingface.co/sentence-transformers
• 文档：https://www.sbert.net
• 许可证：Apache 2.0