Struct Parameters

pub struct Parameters { /* private fields */ }

Parameters

Prefer using default() and set individual parameters as needed.

Implementations

impl Parameters

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

Input truncation (nb of tokens)

pub fn max_length(&self) -> Option<usize>

Input truncation (nb of tokens)

pub fn with_sigmoid(self, sigmoid: bool) -> Self

Apply sigmoid (for re-reanking)

Examples found in repository

examples/rerank.rs (line 10)

fn main() -> gte::util::result::Result<()> {    
    const TOKENIZER_PATH: &str = "models/gte-modernbert-base/tokenizer.json";
    const MODEL_PATH: &str = "models/gte-reranker-modernbert-base/onnx/model.onnx";
    const APPLY_SIGMOID: bool = true;

    let params = gte::params::Parameters::default().with_sigmoid(APPLY_SIGMOID);
    let pipeline = gte::rerank::pipeline::RerankingPipeline::new(TOKENIZER_PATH, &params)?;
    let model = orp::model::Model::new(MODEL_PATH, orp::params::RuntimeParameters::default())?;

    let inputs = gte::rerank::input::TextInput::from_str(&[
        ("what is the capital of China?", "Beijing"),
        ("how to implement quick sort in python?", "Introduction of quick sort"),
        ("how to implement quick sort in python?", "The weather is nice today"),
    ]);

    const EXPECTED_SIMILARITIES: [f32; 3] = if APPLY_SIGMOID { [0.8945664, 0.9213594, 0.15742092] } else { [2.1387, 2.4609, -1.6729] };    
    const EPSILON: f32 = if APPLY_SIGMOID { 0.00001 } else { 0.01 };

    let outputs = model.inference(inputs, &pipeline, &params)?;

    println!("Scores: {:?}", outputs.scores);
    assert!(gte::util::test::is_close_to_a(&outputs.scores.view(), &EXPECTED_SIMILARITIES, EPSILON));

    Ok(())
}

pub fn sigmoid(&self) -> bool

Apply sigmoid (for re-reanking)

pub fn with_output_id(self, id: &str) -> Self

Set output tensor identifier (eg. last_hidden_state)

Examples found in repository

examples/embed-mul-token.rs (line 7)

fn main() -> gte::util::result::Result<()> {    
    const TOKENIZER_PATH: &str = "models/gte-multilingual-base/tokenizer.json";
    const MODEL_PATH: &str = "models/gte-multilingual-base/onnx/model.onnx";

    let params = gte::params::Parameters::default().with_output_id("token_embeddings");
    let pipeline = gte::embed::pipeline::TextEmbeddingPipeline::new(TOKENIZER_PATH, &params)?;
    let model = orp::model::Model::new(MODEL_PATH, orp::params::RuntimeParameters::default())?;
            
    let inputs = gte::embed::input::TextInput::from_str(&[
        "What is the capital of France?", 
        "How to implement quick sort in python?", 
        "Die Hauptstadt von Frankreich ist Paris.",
        "La capital de Francia es París.",
        "London is the capital of the UK",
    ]);

    let outputs = model.inference(inputs, &pipeline, &params)?;
    let distances = gte::util::test::distances_to_first(&outputs);

    println!("Distances: {:?}", distances);     

    Ok(())
}

examples/embed-mul-sentence.rs (line 8)

fn main() -> gte::util::result::Result<()> {    
    const TOKENIZER_PATH: &str = "models/gte-multilingual-base/tokenizer.json";
    const MODEL_PATH: &str = "models/gte-multilingual-base/onnx/model.onnx";

    let params = gte::params::Parameters::default()
        .with_output_id("sentence_embedding")
        .with_mode(gte::embed::output::ExtractorMode::Raw);

    let pipeline = gte::embed::pipeline::TextEmbeddingPipeline::new(TOKENIZER_PATH, &params)?;
    let model = orp::model::Model::new(MODEL_PATH, orp::params::RuntimeParameters::default())?;
            
    let inputs = gte::embed::input::TextInput::from_str(&[
        "What is the capital of France?", 
        "How to implement quick sort in python?", 
        "Die Hauptstadt von Frankreich ist Paris.",
        "La capital de Francia es París.",
        "London is the capital of the UK",
    ]);

    let outputs = model.inference(inputs, &pipeline, &params)?;
    let distances = gte::util::test::distances_to_first(&outputs);

    println!("Distances: {:?}", distances);     

    Ok(())
}

pub fn output_id(&self) -> &OutputId

Output tensor identifier

pub fn with_mode(self, mode: ExtractorMode) -> Self

Set embeddings extraction mode

Examples found in repository

examples/embed-mul-sentence.rs (line 9)

fn main() -> gte::util::result::Result<()> {    
    const TOKENIZER_PATH: &str = "models/gte-multilingual-base/tokenizer.json";
    const MODEL_PATH: &str = "models/gte-multilingual-base/onnx/model.onnx";

    let params = gte::params::Parameters::default()
        .with_output_id("sentence_embedding")
        .with_mode(gte::embed::output::ExtractorMode::Raw);

    let pipeline = gte::embed::pipeline::TextEmbeddingPipeline::new(TOKENIZER_PATH, &params)?;
    let model = orp::model::Model::new(MODEL_PATH, orp::params::RuntimeParameters::default())?;
            
    let inputs = gte::embed::input::TextInput::from_str(&[
        "What is the capital of France?", 
        "How to implement quick sort in python?", 
        "Die Hauptstadt von Frankreich ist Paris.",
        "La capital de Francia es París.",
        "London is the capital of the UK",
    ]);

    let outputs = model.inference(inputs, &pipeline, &params)?;
    let distances = gte::util::test::distances_to_first(&outputs);

    println!("Distances: {:?}", distances);     

    Ok(())
}

pub fn mode(&self) -> ExtractorMode

Embeddings extraction mode

Trait Implementations

impl Default for Parameters

fn default() -> Self

Returns the “default value” for a type.