Struct TextInput

pub struct TextInput {
    pub texts: Vec<String>,
}

Input for text embedding

Fields

texts: Vec<String>

Implementations

impl TextInput

pub fn new(texts: Vec<String>) -> Self

pub fn from_str(texts: &[&str]) -> Self

Examples found in repository

examples/embed-mul-token.rs (lines 11-17)

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 (lines 14-20)

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(())
}

examples/embed.rs (lines 14-19)

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

    let params = gte::params::Parameters::default();
    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 China?", 
        "how to implement quick sort in python?", 
        "Beijing", 
        "sorting algorithms",
    ]);

    const EXPECTED_DISTANCES: [f32; 3] = [0.4289073944091797, 0.7130911254882812, 0.33664554595947266];
    const EPSILON: f32 = 0.000001;

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

    println!("Distances: {:?}", distances); 
    assert!(gte::util::test::is_close_to_a(&distances.view(), &EXPECTED_DISTANCES, EPSILON));

    Ok(())
}

Trait Implementations

impl TextInput<'_> for TextInput

type InputType = String

fn into_encode_input(self) -> Vec<Self::InputType>