Sentence Transformer¶

The TransformersSharp.SentenceTransformer class provides a high-level interface for generating sentence embeddings using pre-trained models from over 10,000 Sentence Similarity models using Hugging Face Transformers library (sentence_transformers). It simplifies the process of embedding text by handling preprocessing, model inference, and tensor conversion.

It is implemented as a Microsoft.AI.Extensions.IEmbeddingGenerator<string, Embedding<float>> and can be used to generate embeddings for sentences or text inputs.

Additionally, the float[] Generate(string input) method is available for generating embeddings for a single input string.

What is a Sentence Transformer?¶

A sentence transformer is designed to generate dense vector representations (embeddings) for input sentences. These embeddings are commonly used for tasks like:

Semantic similarity
Clustering
Information retrieval
Text classification

Key Features of Sentence Transformers:¶

Pre-trained Models: Leverages state-of-the-art models like nomic-ai/nomic-embed-text-v1.5.
High-Dimensional Embeddings: Produces embeddings with hundreds of dimensions for rich semantic representation.
Batch Processing: Supports generating embeddings for multiple sentences in a single call.

Using the SentenceTransformer Class¶

The SentenceTransformer class in TransformersSharp provides methods to generate embeddings for sentences. Below are examples of how to use it.

Simple Usage¶

using TransformersSharp;

var transformer = SentenceTransformer.FromModel("nomic-ai/nomic-embed-text-v1.5", trustRemoteCode: true);
var sentence = "The quick brown fox jumps over the lazy dog.";
float[] embedding = transformer.Generate(sentence);
Console.WriteLine($"Embedding Length: {embedding}");

Generating Embeddings for Sentences¶

using TransformersSharp;

var transformer = SentenceTransformer.FromModel("nomic-ai/nomic-embed-text-v1.5", trustRemoteCode: true);
var sentences = new List<string>
{
    "The quick brown fox jumps over the lazy dog.",
    "Transformers are amazing for natural language processing."
};
var embeddings = await transformer.GenerateAsync(sentences);

foreach (var embedding in embeddings)
{
    Console.WriteLine($"Embedding Length: {embedding.Vector.Length}");
}

Equivalent Python Code:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("nomic-ai/nomic-embed-text-v1.5")
sentences = [
    "The quick brown fox jumps over the lazy dog.",
    "Transformers are amazing for natural language processing."
]
embeddings = model.encode(sentences)

for embedding in embeddings:
    print(f"Embedding Length: {len(embedding)}")

Accessing Embedding Data¶

The SentenceTransformer class provides embeddings as Embedding<float> objects, which contain dense vectors for each input sentence.

Example: Accessing Embedding Data¶

using TransformersSharp;

var transformer = SentenceTransformer.FromModel("nomic-ai/nomic-embed-text-v1.5", trustRemoteCode: true);
var sentences = new List<string>
{
    "The quick brown fox jumps over the lazy dog.",
    "Transformers are amazing for natural language processing."
};
var embeddings = await transformer.GenerateAsync(sentences);

foreach (var embedding in embeddings)
{
    Console.WriteLine($"Vector: {string.Join(", ", embedding.Vector.Take(10))}...");
}

Equivalent Python Code:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("nomic-ai/nomic-embed-text-v1.5")
sentences = [
    "The quick brown fox jumps over the lazy dog.",
    "Transformers are amazing for natural language processing."
]
embeddings = model.encode(sentences)

for embedding in embeddings:
    print(f"Vector: {embedding[:10]}...")