Expand description
ONNX embedder via tract (pure Rust, no PyTorch/Python runtime — brief §8).
Loads a MiniLM-class sentence-embedding ONNX model from a configured path.
Deployment note: model files are customer-supplied artifacts (air-gapped
installs cannot download); the default deployment uses crate::HashEmbedder
and swapping to ONNX is a config change, not a code change (plan D6).
Tokenization is supplied by the model’s Hugging Face tokenizer.json.
Structs§
- Onnx
Embedder - Embedder backed by an ONNX sentence-embedding model.