Struct rust_bert::gpt2::GPT2Generator

pub struct GPT2Generator { /* fields omitted */ }

Language generation model based on the GPT2 architecture

Implementations

impl GPT2Generator

pub fn new(
    generate_config: GenerateConfig
) -> Result<GPT2Generator, RustBertError>

Build a new GPT2Generator

Arguments

• generate_config - GenerateConfig object containing the resource references (model, vocabulary, configuration), generation options and device placement (CPU/GPU)

Example

use rust_bert::gpt2::GPT2Generator;
use rust_bert::pipelines::generation_utils::GenerateConfig;

let generate_config = GenerateConfig {
    max_length: 30,
    do_sample: true,
    num_beams: 5,
    temperature: 1.1,
    num_return_sequences: 3,
    ..Default::default()
};
let gpt2_generator = GPT2Generator::new(generate_config)?;

Trait Implementations

impl LanguageGenerator<GPT2LMHeadModel, Gpt2Vocab, Gpt2Tokenizer> for GPT2Generator

fn generate<S>(
    &self,
    prompt_texts: Option<&[S]>,
    generate_options: Option<GenerateOptions<'_>>
) -> Vec<GeneratedTextOutput> where
    S: AsRef<str> + Sync, 

Generate text based on a vector of promp texts.

fn generate_indices<S>(
    &self,
    prompt_texts: Option<&[S]>,
    generate_options: Option<GenerateOptions<'_>>
) -> Vec<GeneratedIndicesOutput> where
    S: AsRef<str> + Sync, 

Generate token indices without decoding (useful for token-level operations before returning final text or as validation step during training).

fn generate_from_ids_and_past(
    &self,
    input_ids: Tensor,
    attention_mask: Option<Tensor>,
    generate_options: Option<GenerateOptions<'_>>
) -> Vec<GeneratedIndicesOutput>

Generate token indices given a list of indices (useful when the input has been pre-tokenized). Returns a list of output tokens that need to be decoded using a tokenizer.

fn get_tokenizer(&self) -> &TokenizerOption

Returns a reference to the text generator’s tokenizer

fn half(&mut self)

fn float(&mut self)

fn set_device(&mut self, device: Device)