LMSupply.Embedder 0.111.0

LMSupply.Embedder

Local text embedding for .NET with automatic model downloading.

Features

• Zero-config: Models download automatically from HuggingFace
• GPU Acceleration: CUDA, DirectML (Windows), CoreML (macOS)
• Cross-platform: Windows, Linux, macOS
• Simple API: Just 2 lines of code to get started

Quick Start

using LMSupply.Embedder;

// Load the default model
await using var model = await LocalEmbedder.LoadAsync("default");

// Generate embeddings
float[] embedding = await model.EmbedAsync("Hello, world!");
Console.WriteLine($"Dimensions: {embedding.Length}");

Query/Passage Embeddings

Some models (the E5 family, Nomic) are fine-tuned with an asymmetric text-prefix convention — query embeddings and document/passage embeddings need different prefixes for accurate retrieval. EmbedQueryAsync/EmbedPassageAsync apply the right prefix automatically from the model's ModelInfo (a no-op passthrough for models that don't need one):

await using var model = await LocalEmbedder.LoadAsync("multilingual-e5-base");

float[] queryEmbedding = await model.EmbedQueryAsync("what is the capital of France?");
float[] passageEmbedding = await model.EmbedPassageAsync("Paris is the capital of France.");

Batch and Matryoshka-truncated (dimensions:) overloads exist for both, mirroring EmbedAsync.

Available Models

Four standard aliases (LocalEmbedder.LoadAsync("default"), etc.) plus a longer list of models loadable by their explicit short name (LocalEmbedder.LoadAsync("multilingual-e5-base")).

Aliases

Explicit models (by short name)

"Prefix" marks models fine-tuned with the query/passage convention (see Query/Passage Embeddings above) — — means the model needs no prefix and EmbedQueryAsync/EmbedPassageAsync behave as a plain passthrough for it.

GPU Acceleration

Do not add ONNX Runtime packages (Microsoft.ML.OnnxRuntime*): LMSupply provisions the runtime itself, and a second copy conflicts with it. ExecutionProvider.Auto uses CUDA when the CUDA 12 runtime and cuDNN 9 are installed on the machine, CoreML on macOS, and the CPU otherwise. On Windows with an AMD or Intel GPU, ONNX sessions run on the CPU (DirectML was removed in 0.67.0). See GPU acceleration.