Struct JinaCodeEmbedder 

pub struct JinaCodeEmbedder { /* private fields */ }

Jina Code Embeddings 1.5B specialized embedder.

This embedder is optimized for the jina-code-embeddings-1.5b model with:

• Task-specific instruction prefixes (NL2Code, Code2Code, etc.)
• Last-token pooling (required for this decoder-based model)
• Matryoshka dimension truncation (128, 256, 512, 1024, 1536)
• Automatic handling of the 32768 context window

Example

use aurora_semantic::{JinaCodeEmbedder, EmbeddingTask, MatryoshkaDimension};

// Load the model
let embedder = JinaCodeEmbedder::from_directory("./models/jina-code-1.5b")?
    .with_task(EmbeddingTask::NL2Code)
    .with_dimension(MatryoshkaDimension::D512);  // Smaller embeddings

// Generate embeddings for search queries
let query_embedding = embedder.embed("function to parse JSON string")?;

// Generate embeddings for code (use Code2Code task)
let code_embedder = JinaCodeEmbedder::from_directory("./models/jina-code-1.5b")?
    .with_task(EmbeddingTask::Code2Code);
let code_embedding = code_embedder.embed("fn parse_json(s: &str) -> Value { ... }")?;

Implementations

impl JinaCodeEmbedder

pub const DEFAULT_MAX_LENGTH: usize = 32_768usize

Default max sequence length for Jina Code 1.5B.

pub const DEFAULT_DIMENSION: usize = 1_536usize

Default dimension for Jina Code 1.5B.

pub fn from_directory<P: AsRef<Path>>(model_dir: P) -> Result<Self>

Load Jina Code Embeddings 1.5B from a model directory.

The directory should contain:

• model.onnx - The ONNX model file
• tokenizer.json - The HuggingFace tokenizer

pub fn from_onnx_embedder(inner: OnnxEmbedder) -> Self

Create from an existing OnnxEmbedder.

pub fn with_task(self, task: EmbeddingTask) -> Self

Set the embedding task (determines instruction prefix).

pub fn with_dimension(self, dimension: MatryoshkaDimension) -> Self

Set the output dimension (Matryoshka truncation).

Smaller dimensions reduce storage and speed up similarity search with minimal quality loss.

pub fn with_max_length(self, max_length: usize) -> Self

Set the maximum sequence length.

pub fn task(&self) -> EmbeddingTask

Get the current task.

pub fn output_dimension(&self) -> MatryoshkaDimension

Get the output dimension.

pub fn execution_provider(&self) -> &ExecutionProviderInfo

Get information about the execution provider (CPU/GPU).

pub fn is_gpu_accelerated(&self) -> bool

Check if GPU acceleration is being used.

pub fn embed_query(&self, text: &str) -> Result<Vec<f32>>

Embed a search query with the query instruction prefix.

Use this for user queries when searching the index.

Example

let query_embedding = embedder.embed_query("function to parse JSON")?;

pub fn embed_passage(&self, text: &str) -> Result<Vec<f32>>

Embed a code snippet/passage with the passage instruction prefix.

Use this when indexing code to build the search index.

Example

let code_embedding = embedder.embed_passage("fn parse_json(s: &str) -> Value { ... }")?;

pub fn embed_queries(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>>

Embed multiple queries in batch.

pub fn embed_passages(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>>

Embed multiple code passages in batch.

Use this for efficient bulk indexing.

Trait Implementations

impl Embedder for JinaCodeEmbedder

fn embed(&self, text: &str) -> Result<Vec<f32>>

Generate an embedding for a single text.

fn embed_batch(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>>

Generate embeddings for multiple texts in batch.

fn dimension(&self) -> usize

Get the embedding dimension.

fn name(&self) -> &'static str

Get the name of this embedder.

fn max_sequence_length(&self) -> usize

Get the maximum sequence length supported.