Struct OnnxEmbeddingProvider 

pub struct OnnxEmbeddingProvider { /* private fields */ }

ONNX Runtime embedding provider.

Uses all-MiniLM-L6-v2 (384 dimensions) for semantic text embeddings. When the ONNX runtime is not available (no onnx-embeddings feature or missing model file), falls back to a position-aware token-hashing approach that is architecturally compatible with real inference.

Implementations

impl OnnxEmbeddingProvider

pub const DEFAULT_DIMS: usize = 384

Default output dimensionality (all-MiniLM-L6-v2).

pub const DEFAULT_MAX_TOKENS: usize = 128

Default maximum input tokens for code snippets.

pub const MODEL_NAME: &'static str = "all-MiniLM-L6-v2"

Model identifier.

pub fn new(model_path: impl Into<PathBuf>) -> Self

Create a new ONNX provider pointing at the given model path.

If the model file does not exist or the onnx-embeddings feature is disabled, the provider transparently falls back to token-hashing.

pub fn with_config( model_path: impl Into<PathBuf>, dimensions: usize, max_tokens: usize, ) -> Self

Create a provider with custom dimensions and max tokens.

pub fn is_runtime_available(&self) -> bool

Whether the real ONNX runtime is active (vs. fallback).

pub fn model_path(&self) -> &PathBuf

Path to the configured model file.

pub fn max_tokens(&self) -> usize

Maximum input token count.

Trait Implementations

impl EmbeddingProvider for OnnxEmbeddingProvider

fn embed<'life0, 'life1, 'async_trait>( &'life0 self, text: &'life1 str, ) -> Pin<Box<dyn Future<Output = Result<Vec<f32>, EmbeddingError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Embed a single text chunk into a vector.

fn embed_batch<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, texts: &'life1 [&'life2 str], ) -> Pin<Box<dyn Future<Output = Result<Vec<Vec<f32>>, EmbeddingError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Embed a batch of text chunks.

fn dimensions(&self) -> usize

Dimensionality of the output vectors.

fn model_name(&self) -> &str

Name of the embedding model (for metadata tracking).

fn embed_query<'life0, 'life1, 'async_trait>( &'life0 self, query: &'life1 str, ) -> Pin<Box<dyn Future<Output = Result<Vec<f32>, EmbeddingError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Embed a query (asymmetric retrieval lane).

fn warm<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = ()> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Pre-warm the backend: run one throwaway embed so the model graph, runtime thread pool, and allocator are hot before the first real call. Best-effort — errors are swallowed (a backend with nothing to warm, like the Mock, just pays one cheap call). Intended to be called once at daemon startup so the first conversation turn doesn’t pay the model’s first-inference cost (ADR-058 Phase 5.3; mirrors the 6.2 STT pre-warm).