Struct TextEmbedding 

pub struct TextEmbedding {
    pub tokenizer: Tokenizer,
    /* private fields */
}

Rust representation of the TextEmbedding model

Fields

tokenizer: Tokenizer

Implementations

impl TextEmbedding

pub fn try_new( options: InitOptionsWithLength<EmbeddingModel>, ) -> Result<TextEmbedding, Error>

Available on crate feature hf-hub only.

Try to generate a new TextEmbedding Instance

Uses the highest level of Graph optimization

Uses the total number of CPUs available as the number of intra-threads

pub fn try_new_from_user_defined( model: UserDefinedEmbeddingModel, options: InitOptionsUserDefined, ) -> Result<TextEmbedding, Error>

Create a TextEmbedding instance from model files provided by the user.

This can be used for ‘bring your own’ embedding models

pub fn get_default_pooling_method( model_name: &EmbeddingModel, ) -> Option<Pooling>

pub fn get_quantization_mode(model_name: &EmbeddingModel) -> QuantizationMode

Get the quantization mode of the model.

Any models with a Q suffix in their name are quantized models.

Currently only 6 supported models have dynamic quantization:

• Alibaba-NLP/gte-base-en-v1.5
• Alibaba-NLP/gte-large-en-v1.5
• mixedbread-ai/mxbai-embed-large-v1
• nomic-ai/nomic-embed-text-v1.5
• Xenova/all-MiniLM-L12-v2
• Xenova/all-MiniLM-L6-v2

pub fn list_supported_models() -> Vec<ModelInfo<EmbeddingModel>>

Retrieve a list of supported models

pub fn get_model_info( model: &EmbeddingModel, ) -> Result<&ModelInfo<EmbeddingModel>, Error>

Get ModelInfo from EmbeddingModel

pub fn transform<S>( &mut self, texts: impl AsRef<[S]>, batch_size: Option<usize>, ) -> Result<EmbeddingOutput, Error>

where S: AsRef<str> + Send + Sync,

Method to generate an [ort::SessionOutputs] wrapped in a EmbeddingOutput instance, which can be used to extract the embeddings with default or custom methods as well as output key precedence.

Metadata that could be useful for creating the array transformer is returned alongside the EmbeddingOutput instance, such as pooling methods etc.

Note

This is a lower level method than TextEmbedding::embed, and is useful when you need to extract the session outputs in a custom way.

If you want to extract the embeddings directly, use TextEmbedding::embed.

If you want to use the raw session outputs, use EmbeddingOutput::into_raw on the output of this method.

If you want to choose a different export key or customize the way the batch arrays are aggregated, you can define your own array transformer and use it on EmbeddingOutput::export_with_transformer to extract the embeddings with your custom output type.

pub fn embed<S>( &mut self, texts: impl AsRef<[S]>, batch_size: Option<usize>, ) -> Result<Vec<Vec<f32>>, Error>

where S: AsRef<str> + Send + Sync,

Method to generate sentence embeddings for a collection of texts.

Accepts anything that can be referenced as a slice of elements implementing AsRef<str>, such as Vec<String>, Vec<&str>, &[String], or &[&str].

The output is a Vec of Embeddings.

Note

This method is a higher level method than TextEmbedding::transform by utilizing the default output precedence and array transformer for the TextEmbedding model.