Expand description
Zero-dep embedding model catalog.
Mirrors fastembed::EmbeddingModel (44 variants) so the catalog is always
available — even when the embedding-fastembed feature is disabled.
§Backend availability
Each model’s backend support is independent — a model listed below is an
ONNX checkpoint served via embedding-fastembed; the same model family
may also have a Rust-native path on specific platforms:
| Backend (feature) | Models | Platform |
|---|---|---|
embedding-fastembed (ONNX) | All 44 variants below | cross-platform |
embedding-metal (candle) | Qwen3-Embedding, Nomic V2 MoE | macOS (Metal) |
embedding-mlx (MLX) | 13 vanilla-BERT models (see mlx_supported) | macOS (aarch64) |
The MLX provider (MlxEmbeddingProvider, feature embedding-mlx) covers the
vanilla-BERT encoder family: BGE-en-v1.5 (small/base/large), bge-small-zh-v1.5,
all-MiniLM-L6/L12, paraphrase-multilingual-MiniLM, multilingual-e5-small,
Snowflake Arctic (xs/s/m/l) and mxbai-embed-large. Other architectures
(XLM-R, MPNet, NomicBert, GTE NewModel, JinaBert, ModernBERT, Gemma, CLIP)
need their own forward passes. See EmbeddingModel::mlx_supported for the
exclusion list and src/embedding/mlx.rs.
Enums§
- Embedding
Model - Embedding model catalog with metadata for all supported ONNX models.