Struct rust_bert::pipelines::generation::BartGenerator

pub struct BartGenerator { /* fields omitted */ }

Language generation model based on the Bart architecture

Methods

impl BartGenerator

pub fn new(generate_config: GenerateConfig) -> Fallible<BartGenerator>

Build a new BartGenerator

Arguments

• vocab_path - Path to the model vocabulary, expected to have a structure following the Transformers library convention
• merges_path - Path to the bpe merges, expected to have a structure following the Transformers library convention
• config_path - Path to the model configuration, expected to have a structure following the Transformers library convention
• weights_path - Path to the model weight files. These need to be converted form the .bin to .ot format using the utility script provided.
• device - Device to run the model on, e.g. Device::Cpu or Device::Cuda(0)

Example

use rust_bert::pipelines::generation::{GenerateConfig, BartGenerator};
let device = Device::cuda_if_available();
let generate_config = GenerateConfig {
   max_length: 30,
   do_sample: true,
   num_beams: 5,
   temperature: 1.1,
   num_return_sequences: 3,
   ..Default::default()
};
let bart_generator = BartGenerator::new(generate_config)?;

Trait Implementations

impl LanguageGenerator<BartForConditionalGeneration, RobertaVocab, RobertaTokenizer> for BartGenerator

fn generate(
    &mut self,
    prompt_texts: Option<Vec<&str>>,
    attention_mask: Option<Tensor>
) -> Vec<String>

Generate text based on a vector of promp texts.